Blame - docs/dyn/ml_v1.projects.jobs.html - platform/external/python/google-api-python-client

2017-03-13 12:12:03 -0400

[diff] [blame]

76

<h2>Instance Methods</h2>

77

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

78

<code><a href="#cancel">cancel(name, body=None, x__xgafv=None)</a></code></p>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

79

<p class="firstline">Cancels a running job.</p>

80

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

81

<code><a href="#create">create(parent, body=None, x__xgafv=None)</a></code></p>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

82

<p class="firstline">Creates a training or a batch prediction job.</p>

83

Thomas Coffee

2f24537

2017-03-27 10:39:26 -0700

[diff] [blame]

84

<code><a href="#get">get(name, x__xgafv=None)</a></code></p>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

85

<p class="firstline">Describes a job.</p>

86

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

87

<code><a href="#getIamPolicy">getIamPolicy(resource, options_requestedPolicyVersion=None, x__xgafv=None)</a></code></p>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

88

<p class="firstline">Gets the access control policy for a resource.</p>

89

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

90

<code><a href="#list">list(parent, filter=None, pageToken=None, pageSize=None, x__xgafv=None)</a></code></p>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

91

<p class="firstline">Lists the jobs in the project.</p>

92

93

<code><a href="#list_next">list_next(previous_request, previous_response)</a></code></p>

94

<p class="firstline">Retrieves the next page of results.</p>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

95

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

96

<code><a href="#patch">patch(name, body=None, updateMask=None, x__xgafv=None)</a></code></p>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

97

<p class="firstline">Updates a specific job resource.</p>

98

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

99

<code><a href="#setIamPolicy">setIamPolicy(resource, body=None, x__xgafv=None)</a></code></p>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

100

<p class="firstline">Sets the access control policy on the specified resource. Replaces any</p>

101

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

102

<code><a href="#testIamPermissions">testIamPermissions(resource, body=None, x__xgafv=None)</a></code></p>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

103

<p class="firstline">Returns permissions that a caller has on the specified resource.</p>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

104

<h3>Method Details</h3>

105

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

106

<code class="details" id="cancel">cancel(name, body=None, x__xgafv=None)</code>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

107

<pre>Cancels a running job.

108

109

Args:

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

110

name: string, Required. The name of the job to cancel. (required)

111

body: object, The request body.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

112

The object takes the form of:

113

114

{ # Request message for the CancelJob method.

115

}

116

117

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

124

125

{ # A generic empty message that you can re-use to avoid defining duplicated

126

# empty messages in your APIs. A typical example is to use it as the request

127

# or the response type of an API method. For instance:

128

#

129

# service Foo {

130

# rpc Bar(google.protobuf.Empty) returns (google.protobuf.Empty);

131

# }

132

#

133

# The JSON representation for `Empty` is empty JSON object `{}`.

}</pre>

</div>

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

138

<code class="details" id="create">create(parent, body=None, x__xgafv=None)</code>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

139

<pre>Creates a training or a batch prediction job.

140

141

Args:

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

142

parent: string, Required. The project name. (required)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

143

body: object, The request body.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

144

The object takes the form of:

145

146

{ # Represents a training or prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

147

"createTime": "A String", # Output only. When the job was created.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

148

"predictionInput": { # Represents input parameters for a prediction job. # Input parameters to create a prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

149

"dataFormat": "A String", # Required. The format of the input data files.

150

"outputPath": "A String", # Required. The output Google Cloud Storage location.

151

"inputPaths": [ # Required. The Cloud Storage location of the input data files. May contain

152

# <a href="/storage/docs/gsutil/addlhelp/WildcardNames">wildcards</a>.

153

"A String",

154

],

155

"region": "A String", # Required. The Google Compute Engine region to run the prediction job in.

156

# See the <a href="/ml-engine/docs/tensorflow/regions">available regions</a>

157

# for AI Platform services.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

158

"versionName": "A String", # Use this field if you want to specify a version of the model to use. The

159

# string is formatted the same way as `model_version`, with the addition

160

# of the version information:

161

#

162

# `"projects/YOUR_PROJECT/models/YOUR_MODEL/versions/YOUR_VERSION"`

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

163

"uri": "A String", # Use this field if you want to specify a Google Cloud Storage path for

164

# the model to use.

165

"outputDataFormat": "A String", # Optional. Format of the output data files, defaults to JSON.

166

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for this batch

167

# prediction. If not set, AI Platform will pick the runtime version used

168

# during the CreateVersion request for this model version, or choose the

169

# latest stable version when model version information is not available

170

# such as when the model is specified by uri.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

171

"modelName": "A String", # Use this field if you want to use the default version for the specified

172

# model. The string must use the following format:

173

#

174

# `"projects/YOUR_PROJECT/models/YOUR_MODEL"`

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

175

"signatureName": "A String", # Optional. The name of the signature defined in the SavedModel to use for

176

# this job. Please refer to

177

# [SavedModel](https://tensorflow.github.io/serving/serving_basic.html)

178

# for information about how to use signatures.

179

#

180

# Defaults to

181

# [DEFAULT_SERVING_SIGNATURE_DEF_KEY](https://www.tensorflow.org/api_docs/python/tf/saved_model/signature_constants)

182

# , which is "serving_default".

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

183

"batchSize": "A String", # Optional. Number of records per batch, defaults to 64.

184

# The service will buffer batch_size number of records in memory before

185

# invoking one Tensorflow prediction call internally. So take the record

186

# size and memory available into consideration when setting this parameter.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

187

"maxWorkerCount": "A String", # Optional. The maximum number of workers to be used for parallel processing.

188

# Defaults to 10 if not specified.

189

},

190

"labels": { # Optional. One or more labels that you can add, to organize your jobs.

191

# Each label is a key-value pair, where both the key and the value are

192

# arbitrary strings that you supply.

193

# For more information, see the documentation on

194

# <a href="/ml-engine/docs/tensorflow/resource-labels">using labels</a>.

195

"a_key": "A String",

196

},

197

"endTime": "A String", # Output only. When the job processing was completed.

198

"trainingOutput": { # Represents results of a training job. Output only. # The current training job result.

199

"trials": [ # Results for individual Hyperparameter trials.

200

# Only set for hyperparameter tuning jobs.

201

{ # Represents the result of a single hyperparameter tuning trial from a

202

# training job. The TrainingOutput object that is returned on successful

203

# completion of a training job with hyperparameter tuning includes a list

204

# of HyperparameterOutput objects, one for each successful trial.

205

"endTime": "A String", # Output only. End time for the trial.

206

"finalMetric": { # An observed value of a metric. # The final objective metric seen for this trial.

207

"trainingStep": "A String", # The global training step for this metric.

208

"objectiveValue": 3.14, # The objective value at this training step.

209

},

210

"hyperparameters": { # The hyperparameters given to this trial.

211

"a_key": "A String",

212

},

213

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

214

# Only set for trials of built-in algorithms jobs that have succeeded.

215

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

216

# saves the trained model. Only set for successful jobs that don't use

217

# hyperparameter tuning.

218

"framework": "A String", # Framework on which the built-in algorithm was trained.

219

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

220

# trained.

221

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

222

},

223

"startTime": "A String", # Output only. Start time for the trial.

224

"allMetrics": [ # All recorded object metrics for this trial. This field is not currently

225

# populated.

226

{ # An observed value of a metric.

227

"trainingStep": "A String", # The global training step for this metric.

228

"objectiveValue": 3.14, # The objective value at this training step.

229

},

230

],

231

"trialId": "A String", # The trial id for these results.

232

"isTrialStoppedEarly": True or False, # True if the trial is stopped early.

233

"state": "A String", # Output only. The detailed state of the trial.

234

},

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

235

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

236

"completedTrialCount": "A String", # The number of hyperparameter tuning trials that completed successfully.

237

# Only set for hyperparameter tuning jobs.

238

"isHyperparameterTuningJob": True or False, # Whether this job is a hyperparameter tuning job.

239

"isBuiltInAlgorithmJob": True or False, # Whether this job is a built-in Algorithm job.

240

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

241

# Only set for built-in algorithms jobs.

242

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

243

# saves the trained model. Only set for successful jobs that don't use

244

# hyperparameter tuning.

245

"framework": "A String", # Framework on which the built-in algorithm was trained.

246

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

247

# trained.

248

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

249

},

250

"consumedMLUnits": 3.14, # The amount of ML units consumed by the job.

251

"hyperparameterMetricTag": "A String", # The TensorFlow summary tag name used for optimizing hyperparameter tuning

252

# trials. See

253

# [`HyperparameterSpec.hyperparameterMetricTag`](#HyperparameterSpec.FIELDS.hyperparameter_metric_tag)

254

# for more information. Only set for hyperparameter tuning jobs.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

255

},

256

"errorMessage": "A String", # Output only. The details of a failure or a cancellation.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

257

"predictionOutput": { # Represents results of a prediction job. # The current prediction job result.

258

"errorCount": "A String", # The number of data instances which resulted in errors.

259

"nodeHours": 3.14, # Node hours used by the batch prediction job.

260

"outputPath": "A String", # The output Google Cloud Storage location provided at the job creation time.

261

"predictionCount": "A String", # The number of generated predictions.

262

},

263

"startTime": "A String", # Output only. When the job processing was started.

264

"state": "A String", # Output only. The detailed state of a job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

265

"trainingInput": { # Represents input parameters for a training job. When using the gcloud command # Input parameters to create a training job.

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

266

# to submit your training job, you can specify the input parameters as

267

# command-line arguments and/or in a YAML configuration file referenced from

268

# the --config command-line argument. For details, see the guide to [submitting

269

# a training job](/ai-platform/training/docs/training-jobs).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

270

"serviceAccount": "A String", # Optional. Specifies the service account for workload run-as account.

271

# Users submitting jobs must have act-as permission on this run-as account.

272

# If not specified, then CMLE P4SA will be used by default.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

273

"workerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for workers.

274

#

275

# You should only set `workerConfig.acceleratorConfig` if `workerType` is set

276

# to a Compute Engine machine type. [Learn about restrictions on accelerator

277

# configurations for

278

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

279

#

280

# Set `workerConfig.imageUri` only if you build a custom image for your

281

# worker. If `workerConfig.imageUri` has not been set, AI Platform uses

282

# the value of `masterConfig.imageUri`. Learn more about [configuring custom

283

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

284

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

285

# [Learn about restrictions on accelerator configurations for

286

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

287

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

288

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

289

# [accelerators for online

290

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

291

"type": "A String", # The type of accelerator to use.

292

"count": "A String", # The number of accelerators to attach to each machine running the job.

293

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

294

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

295

# Registry. Learn more about [configuring custom

296

# containers](/ai-platform/training/docs/distributed-training-containers).

297

"containerArgs": [ # Arguments to the entrypoint command.

298

# The following rules apply for container_command and container_args:

299

# - If you do not supply command or args:

300

# The defaults defined in the Docker image are used.

301

# - If you supply a command but no args:

302

# The default EntryPoint and the default Cmd defined in the Docker image

303

# are ignored. Your command is run without any arguments.

304

# - If you supply only args:

305

# The default Entrypoint defined in the Docker image is run with the args

306

# that you supplied.

307

# - If you supply a command and args:

308

# The default Entrypoint and the default Cmd defined in the Docker image

309

# are ignored. Your command is run with your args.

310

# It cannot be set if custom container image is

311

# not provided.

312

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

313

# both cannot be set at the same time.

314

"A String",

315

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

316

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

317

# the one used in the custom container. This field is required if the replica

318

# is a TPU worker that uses a custom container. Otherwise, do not specify

319

# this field. This must be a [runtime version that currently supports

320

# training with

321

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

322

#

323

# Note that the version of TensorFlow included in a runtime version may

324

# differ from the numbering of the runtime version itself, because it may

325

# have a different [patch

326

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

327

# In this field, you must specify the runtime version (TensorFlow minor

328

# version). For example, if your custom container runs TensorFlow `1.x.y`,

329

# specify `1.x`.

330

"containerCommand": [ # The command with which the replica's custom container is run.

331

# If provided, it will override default ENTRYPOINT of the docker image.

332

# If not provided, the docker image's ENTRYPOINT is used.

333

# It cannot be set if custom container image is

334

# not provided.

335

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

336

# both cannot be set at the same time.

337

"A String",

338

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

339

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

340

"useChiefInTfConfig": True or False, # Optional. Use `chief` instead of `master` in the `TF_CONFIG` environment

341

# variable when training with a custom container. Defaults to `false`. [Learn

342

# more about this

343

# field.](/ai-platform/training/docs/distributed-training-details#chief-versus-master)

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

344

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

345

# This field has no effect for training jobs that don't use a custom

346

# container.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

347

"masterType": "A String", # Optional. Specifies the type of virtual machine to use for your training

348

# job's master worker. You must specify this field when `scaleTier` is set to

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

349

# `CUSTOM`.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

350

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

351

# You can use certain Compute Engine machine types directly in this field.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

352

# The following types are supported:

353

#

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

# - `n1-standard-4`

# - `n1-standard-8`

# - `n1-standard-16`

# - `n1-standard-32`

# - `n1-standard-64`

# - `n1-standard-96`

# - `n1-highmem-2`

# - `n1-highmem-4`

# - `n1-highmem-8`

# - `n1-highmem-16`

# - `n1-highmem-32`

# - `n1-highmem-64`

# - `n1-highmem-96`

# - `n1-highcpu-16`

# - `n1-highcpu-32`

# - `n1-highcpu-64`

# - `n1-highcpu-96`

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

372

# Learn more about [using Compute Engine machine

373

# types](/ml-engine/docs/machine-types#compute-engine-machine-types).

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

374

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

375

# Alternatively, you can use the following legacy machine types:

#

# - `standard`

# - `large_model`

# - `complex_model_s`

380

# - `complex_model_m`

381

# - `complex_model_l`

382

# - `standard_gpu`

383

# - `complex_model_m_gpu`

384

# - `complex_model_l_gpu`

385

# - `standard_p100`

386

# - `complex_model_m_p100`

387

# - `standard_v100`

388

# - `large_model_v100`

389

# - `complex_model_m_v100`

390

# - `complex_model_l_v100`

391

#

392

# Learn more about [using legacy machine

393

# types](/ml-engine/docs/machine-types#legacy-machine-types).

394

#

395

# Finally, if you want to use a TPU for training, specify `cloud_tpu` in this

396

# field. Learn more about the [special configuration options for training

397

# with

398

# TPUs](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

399

"parameterServerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for parameter servers.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

400

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

401

# You should only set `parameterServerConfig.acceleratorConfig` if

402

# `parameterServerType` is set to a Compute Engine machine type. [Learn

403

# about restrictions on accelerator configurations for

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

404

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

405

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

406

# Set `parameterServerConfig.imageUri` only if you build a custom image for

407

# your parameter server. If `parameterServerConfig.imageUri` has not been

408

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

409

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

410

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

411

# [Learn about restrictions on accelerator configurations for

412

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

413

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

414

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

415

# [accelerators for online

416

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

417

"type": "A String", # The type of accelerator to use.

418

"count": "A String", # The number of accelerators to attach to each machine running the job.

419

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

420

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

421

# Registry. Learn more about [configuring custom

422

# containers](/ai-platform/training/docs/distributed-training-containers).

423

"containerArgs": [ # Arguments to the entrypoint command.

424

# The following rules apply for container_command and container_args:

425

# - If you do not supply command or args:

426

# The defaults defined in the Docker image are used.

427

# - If you supply a command but no args:

428

# The default EntryPoint and the default Cmd defined in the Docker image

429

# are ignored. Your command is run without any arguments.

430

# - If you supply only args:

431

# The default Entrypoint defined in the Docker image is run with the args

432

# that you supplied.

433

# - If you supply a command and args:

434

# The default Entrypoint and the default Cmd defined in the Docker image

435

# are ignored. Your command is run with your args.

436

# It cannot be set if custom container image is

437

# not provided.

438

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

439

# both cannot be set at the same time.

440

"A String",

441

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

442

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

443

# the one used in the custom container. This field is required if the replica

444

# is a TPU worker that uses a custom container. Otherwise, do not specify

445

# this field. This must be a [runtime version that currently supports

446

# training with

447

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

448

#

449

# Note that the version of TensorFlow included in a runtime version may

450

# differ from the numbering of the runtime version itself, because it may

451

# have a different [patch

452

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

453

# In this field, you must specify the runtime version (TensorFlow minor

454

# version). For example, if your custom container runs TensorFlow `1.x.y`,

455

# specify `1.x`.

456

"containerCommand": [ # The command with which the replica's custom container is run.

457

# If provided, it will override default ENTRYPOINT of the docker image.

458

# If not provided, the docker image's ENTRYPOINT is used.

459

# It cannot be set if custom container image is

460

# not provided.

461

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

462

# both cannot be set at the same time.

463

"A String",

464

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

465

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

466

"region": "A String", # Required. The region to run the training job in. See the [available

467

# regions](/ai-platform/training/docs/regions) for AI Platform Training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

468

"jobDir": "A String", # Optional. A Google Cloud Storage path in which to store training outputs

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

469

# and other data needed for training. This path is passed to your TensorFlow

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

470

# program as the '--job-dir' command-line argument. The benefit of specifying

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

471

# this field is that Cloud ML validates the path for use in training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

472

"pythonVersion": "A String", # Optional. The version of Python used in training. You must either specify

473

# this field or specify `masterConfig.imageUri`.

474

#

475

# The following Python versions are available:

476

#

477

# * Python '3.7' is available when `runtime_version` is set to '1.15' or

478

# later.

479

# * Python '3.5' is available when `runtime_version` is set to a version

480

# from '1.4' to '1.14'.

481

# * Python '2.7' is available when `runtime_version` is set to '1.15' or

482

# earlier.

483

#

484

# Read more about the Python versions available for [each runtime

485

# version](/ml-engine/docs/runtime-version-list).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

486

"hyperparameters": { # Represents a set of hyperparameters to optimize. # Optional. The set of Hyperparameters to tune.

487

"maxTrials": 42, # Optional. How many training trials should be attempted to optimize

488

# the specified hyperparameters.

489

#

490

# Defaults to one.

491

"enableTrialEarlyStopping": True or False, # Optional. Indicates if the hyperparameter tuning job enables auto trial

492

# early stopping.

493

"params": [ # Required. The set of parameters to tune.

494

{ # Represents a single hyperparameter to optimize.

495

"minValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

496

# should be unset if type is `CATEGORICAL`. This value should be integers if

497

# type is INTEGER.

498

"categoricalValues": [ # Required if type is `CATEGORICAL`. The list of possible categories.

499

"A String",

500

],

501

"scaleType": "A String", # Optional. How the parameter should be scaled to the hypercube.

502

# Leave unset for categorical parameters.

503

# Some kind of scaling is strongly recommended for real or integral

504

# parameters (e.g., `UNIT_LINEAR_SCALE`).

505

"discreteValues": [ # Required if type is `DISCRETE`.

506

# A list of feasible points.

507

# The list should be in strictly increasing order. For instance, this

508

# parameter might have possible settings of 1.5, 2.5, and 4.0. This list

509

# should not contain more than 1,000 values.

510

3.14,

511

],

512

"type": "A String", # Required. The type of the parameter.

513

"maxValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

514

# should be unset if type is `CATEGORICAL`. This value should be integers if

515

# type is `INTEGER`.

516

"parameterName": "A String", # Required. The parameter name must be unique amongst all ParameterConfigs in

517

# a HyperparameterSpec message. E.g., "learning_rate".

518

},

519

],

520

"maxFailedTrials": 42, # Optional. The number of failed trials that need to be seen before failing

521

# the hyperparameter tuning job. You can specify this field to override the

522

# default failing criteria for AI Platform hyperparameter tuning jobs.

523

#

524

# Defaults to zero, which means the service decides when a hyperparameter

525

# job should fail.

526

"hyperparameterMetricTag": "A String", # Optional. The TensorFlow summary tag name to use for optimizing trials. For

527

# current versions of TensorFlow, this tag name should exactly match what is

528

# shown in TensorBoard, including all scopes. For versions of TensorFlow

529

# prior to 0.12, this should be only the tag passed to tf.Summary.

530

# By default, "training/hptuning/metric" will be used.

531

"resumePreviousJobId": "A String", # Optional. The prior hyperparameter tuning job id that users hope to

532

# continue with. The job id will be used to find the corresponding vizier

533

# study guid and resume the study.

534

"goal": "A String", # Required. The type of goal to use for tuning. Available types are

535

# `MAXIMIZE` and `MINIMIZE`.

536

#

537

# Defaults to `MAXIMIZE`.

538

"algorithm": "A String", # Optional. The search algorithm specified for the hyperparameter

539

# tuning job.

540

# Uses the default AI Platform hyperparameter tuning

541

# algorithm if unspecified.

542

"maxParallelTrials": 42, # Optional. The number of training trials to run concurrently.

543

# You can reduce the time it takes to perform hyperparameter tuning by adding

544

# trials in parallel. However, each trail only benefits from the information

545

# gained in completed trials. That means that a trial does not get access to

546

# the results of trials running at the same time, which could reduce the

547

# quality of the overall optimization.

548

#

549

# Each trial will use the same scale tier and machine types.

#

# Defaults to one.

},

"evaluatorType": "A String", # Optional. Specifies the type of virtual machine to use for your training

554

# job's evaluator nodes.

555

#

556

# The supported values are the same as those described in the entry for

557

# `masterType`.

558

#

559

# This value must be consistent with the category of machine type that

560

# `masterType` uses. In other words, both must be Compute Engine machine

561

# types or both must be legacy machine types.

562

#

563

# This value must be present when `scaleTier` is set to `CUSTOM` and

564

# `evaluatorCount` is greater than zero.

565

"network": "A String", # Optional. The full name of the Google Compute Engine

566

# [network](/compute/docs/networks-and-firewalls#networks) to which the Job

567

# is peered. For example, projects/12345/global/networks/myVPC. Format is of

568

# the form projects/{project}/global/networks/{network}. Where {project} is a

569

# project number, as in '12345', and {network} is network name.".

570

#

571

# Private services access must already be configured for the network. If left

572

# unspecified, the Job is not peered with any network. Learn more -

573

# Connecting Job to user network over private

574

# IP.

575

"parameterServerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

576

# job's parameter server.

577

#

578

# The supported values are the same as those described in the entry for

579

# `master_type`.

580

#

581

# This value must be consistent with the category of machine type that

582

# `masterType` uses. In other words, both must be Compute Engine machine

583

# types or both must be legacy machine types.

584

#

585

# This value must be present when `scaleTier` is set to `CUSTOM` and

586

# `parameter_server_count` is greater than zero.

587

"workerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

588

# job's worker nodes.

589

#

590

# The supported values are the same as those described in the entry for

591

# `masterType`.

592

#

593

# This value must be consistent with the category of machine type that

594

# `masterType` uses. In other words, both must be Compute Engine machine

595

# types or both must be legacy machine types.

596

#

597

# If you use `cloud_tpu` for this value, see special instructions for

598

# [configuring a custom TPU

599

# machine](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

600

#

601

# This value must be present when `scaleTier` is set to `CUSTOM` and

602

# `workerCount` is greater than zero.

603

"masterConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for your master worker.

604

#

605

# You should only set `masterConfig.acceleratorConfig` if `masterType` is set

606

# to a Compute Engine machine type. Learn about [restrictions on accelerator

607

# configurations for

608

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

609

#

610

# Set `masterConfig.imageUri` only if you build a custom image. Only one of

611

# `masterConfig.imageUri` and `runtimeVersion` should be set. Learn more

612

# about [configuring custom

613

# containers](/ai-platform/training/docs/distributed-training-containers).

614

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

615

# [Learn about restrictions on accelerator configurations for

616

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

617

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

618

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

619

# [accelerators for online

620

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

621

"type": "A String", # The type of accelerator to use.

622

"count": "A String", # The number of accelerators to attach to each machine running the job.

623

},

624

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

625

# Registry. Learn more about [configuring custom

626

# containers](/ai-platform/training/docs/distributed-training-containers).

627

"containerArgs": [ # Arguments to the entrypoint command.

628

# The following rules apply for container_command and container_args:

629

# - If you do not supply command or args:

630

# The defaults defined in the Docker image are used.

631

# - If you supply a command but no args:

632

# The default EntryPoint and the default Cmd defined in the Docker image

633

# are ignored. Your command is run without any arguments.

634

# - If you supply only args:

635

# The default Entrypoint defined in the Docker image is run with the args

636

# that you supplied.

637

# - If you supply a command and args:

638

# The default Entrypoint and the default Cmd defined in the Docker image

639

# are ignored. Your command is run with your args.

640

# It cannot be set if custom container image is

641

# not provided.

642

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

643

# both cannot be set at the same time.

644

"A String",

645

],

646

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

647

# the one used in the custom container. This field is required if the replica

648

# is a TPU worker that uses a custom container. Otherwise, do not specify

649

# this field. This must be a [runtime version that currently supports

650

# training with

651

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

652

#

653

# Note that the version of TensorFlow included in a runtime version may

654

# differ from the numbering of the runtime version itself, because it may

655

# have a different [patch

656

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

657

# In this field, you must specify the runtime version (TensorFlow minor

658

# version). For example, if your custom container runs TensorFlow `1.x.y`,

659

# specify `1.x`.

660

"containerCommand": [ # The command with which the replica's custom container is run.

661

# If provided, it will override default ENTRYPOINT of the docker image.

662

# If not provided, the docker image's ENTRYPOINT is used.

663

# It cannot be set if custom container image is

664

# not provided.

665

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

666

# both cannot be set at the same time.

"A String",

],

},

"evaluatorCount": "A String", # Optional. The number of evaluator replicas to use for the training job.

671

# Each replica in the cluster will be of the type specified in

672

# `evaluator_type`.

673

#

674

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

675

# set this value, you must also set `evaluator_type`.

676

#

677

# The default value is zero.

678

"args": [ # Optional. Command-line arguments passed to the training application when it

679

# starts. If your job uses a custom container, then the arguments are passed

680

# to the container's <a class="external" target="_blank"

681

# href="https://docs.docker.com/engine/reference/builder/#entrypoint">

682

# `ENTRYPOINT`</a> command.

683

"A String",

684

],

685

"pythonModule": "A String", # Required. The Python module name to run after installing the packages.

686

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for training. You must

687

# either specify this field or specify `masterConfig.imageUri`.

688

#

689

# For more information, see the [runtime version

690

# list](/ai-platform/training/docs/runtime-version-list) and learn [how to

691

# manage runtime versions](/ai-platform/training/docs/versioning).

692

"parameterServerCount": "A String", # Optional. The number of parameter server replicas to use for the training

693

# job. Each replica in the cluster will be of the type specified in

694

# `parameter_server_type`.

695

#

696

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

697

# set this value, you must also set `parameter_server_type`.

698

#

699

# The default value is zero.

700

"evaluatorConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for evaluators.

701

#

702

# You should only set `evaluatorConfig.acceleratorConfig` if

703

# `evaluatorType` is set to a Compute Engine machine type. [Learn

704

# about restrictions on accelerator configurations for

705

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

706

#

707

# Set `evaluatorConfig.imageUri` only if you build a custom image for

708

# your evaluator. If `evaluatorConfig.imageUri` has not been

709

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

710

# containers](/ai-platform/training/docs/distributed-training-containers).

711

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

712

# [Learn about restrictions on accelerator configurations for

713

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

714

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

715

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

716

# [accelerators for online

717

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

718

"type": "A String", # The type of accelerator to use.

719

"count": "A String", # The number of accelerators to attach to each machine running the job.

720

},

721

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

722

# Registry. Learn more about [configuring custom

723

# containers](/ai-platform/training/docs/distributed-training-containers).

724

"containerArgs": [ # Arguments to the entrypoint command.

725

# The following rules apply for container_command and container_args:

726

# - If you do not supply command or args:

727

# The defaults defined in the Docker image are used.

728

# - If you supply a command but no args:

729

# The default EntryPoint and the default Cmd defined in the Docker image

730

# are ignored. Your command is run without any arguments.

731

# - If you supply only args:

732

# The default Entrypoint defined in the Docker image is run with the args

733

# that you supplied.

734

# - If you supply a command and args:

735

# The default Entrypoint and the default Cmd defined in the Docker image

736

# are ignored. Your command is run with your args.

737

# It cannot be set if custom container image is

738

# not provided.

739

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

740

# both cannot be set at the same time.

741

"A String",

742

],

743

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

744

# the one used in the custom container. This field is required if the replica

745

# is a TPU worker that uses a custom container. Otherwise, do not specify

746

# this field. This must be a [runtime version that currently supports

747

# training with

748

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

749

#

750

# Note that the version of TensorFlow included in a runtime version may

751

# differ from the numbering of the runtime version itself, because it may

752

# have a different [patch

753

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

754

# In this field, you must specify the runtime version (TensorFlow minor

755

# version). For example, if your custom container runs TensorFlow `1.x.y`,

756

# specify `1.x`.

757

"containerCommand": [ # The command with which the replica's custom container is run.

758

# If provided, it will override default ENTRYPOINT of the docker image.

759

# If not provided, the docker image's ENTRYPOINT is used.

760

# It cannot be set if custom container image is

761

# not provided.

762

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

763

# both cannot be set at the same time.

"A String",

],

},

"encryptionConfig": { # Represents a custom encryption key configuration that can be applied to # Optional. Options for using customer-managed encryption keys (CMEK) to

768

# protect resources created by a training job, instead of using Google's

769

# default encryption. If this is set, then all resources created by the

770

# training job will be encrypted with the customer-managed encryption key

771

# that you specify.

772

#

773

# [Learn how and when to use CMEK with AI Platform

774

# Training](/ai-platform/training/docs/cmek).

775

# a resource.

776

"kmsKeyName": "A String", # The Cloud KMS resource identifier of the customer-managed encryption key

777

# used to protect a resource, such as a training job. It has the following

778

# format:

779

# `projects/{PROJECT_ID}/locations/{REGION}/keyRings/{KEY_RING_NAME}/cryptoKeys/{KEY_NAME}`

780

},

781

"workerCount": "A String", # Optional. The number of worker replicas to use for the training job. Each

782

# replica in the cluster will be of the type specified in `worker_type`.

783

#

784

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

785

# set this value, you must also set `worker_type`.

786

#

787

# The default value is zero.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

788

"scheduling": { # All parameters related to scheduling of training jobs. # Optional. Scheduling options for a training job.

789

"maxWaitTime": "A String",

790

"maxRunningTime": "A String", # Optional. The maximum job running time, expressed in seconds. The field can

791

# contain up to nine fractional digits, terminated by `s`. If not specified,

792

# this field defaults to `604800s` (seven days).

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

793

#

794

# If the training job is still running after this duration, AI Platform

795

# Training cancels it.

796

#

797

# For example, if you want to ensure your job runs for no more than 2 hours,

798

# set this field to `7200s` (2 hours * 60 minutes / hour * 60 seconds /

799

# minute).

800

#

801

# If you submit your training job using the `gcloud` tool, you can [provide

802

# this field in a `config.yaml`

803

# file](/ai-platform/training/docs/training-jobs#formatting_your_configuration_parameters).

# For example:

#

# ```yaml

# trainingInput:

# ...

# scheduling:

# maxRunningTime: 7200s

811

# ...

812

# ```

813

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

814

"scaleTier": "A String", # Required. Specifies the machine types, the number of replicas for workers

815

# and parameter servers.

816

"packageUris": [ # Required. The Google Cloud Storage location of the packages with

817

# the training program and any additional dependencies.

818

# The maximum number of package URIs is 100.

819

"A String",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

820

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

821

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

822

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

823

# prevent simultaneous updates of a job from overwriting each other.

824

# It is strongly suggested that systems make use of the `etag` in the

825

# read-modify-write cycle to perform job updates in order to avoid race

826

# conditions: An `etag` is returned in the response to `GetJob`, and

827

# systems are expected to put that etag in the request to `UpdateJob` to

828

# ensure that their change will be applied to the same version of the job.

829

"jobId": "A String", # Required. The user-specified id of the job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

830

}

831

832

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

839

840

{ # Represents a training or prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

841

"createTime": "A String", # Output only. When the job was created.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

842

"predictionInput": { # Represents input parameters for a prediction job. # Input parameters to create a prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

843

"dataFormat": "A String", # Required. The format of the input data files.

844

"outputPath": "A String", # Required. The output Google Cloud Storage location.

845

"inputPaths": [ # Required. The Cloud Storage location of the input data files. May contain

846

# <a href="/storage/docs/gsutil/addlhelp/WildcardNames">wildcards</a>.

847

"A String",

848

],

849

"region": "A String", # Required. The Google Compute Engine region to run the prediction job in.

850

# See the <a href="/ml-engine/docs/tensorflow/regions">available regions</a>

851

# for AI Platform services.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

852

"versionName": "A String", # Use this field if you want to specify a version of the model to use. The

853

# string is formatted the same way as `model_version`, with the addition

854

# of the version information:

855

#

856

# `"projects/YOUR_PROJECT/models/YOUR_MODEL/versions/YOUR_VERSION"`

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

857

"uri": "A String", # Use this field if you want to specify a Google Cloud Storage path for

858

# the model to use.

859

"outputDataFormat": "A String", # Optional. Format of the output data files, defaults to JSON.

860

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for this batch

861

# prediction. If not set, AI Platform will pick the runtime version used

862

# during the CreateVersion request for this model version, or choose the

863

# latest stable version when model version information is not available

864

# such as when the model is specified by uri.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

865

"modelName": "A String", # Use this field if you want to use the default version for the specified

866

# model. The string must use the following format:

867

#

868

# `"projects/YOUR_PROJECT/models/YOUR_MODEL"`

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

869

"signatureName": "A String", # Optional. The name of the signature defined in the SavedModel to use for

870

# this job. Please refer to

871

# [SavedModel](https://tensorflow.github.io/serving/serving_basic.html)

872

# for information about how to use signatures.

873

#

874

# Defaults to

875

# [DEFAULT_SERVING_SIGNATURE_DEF_KEY](https://www.tensorflow.org/api_docs/python/tf/saved_model/signature_constants)

876

# , which is "serving_default".

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

877

"batchSize": "A String", # Optional. Number of records per batch, defaults to 64.

878

# The service will buffer batch_size number of records in memory before

879

# invoking one Tensorflow prediction call internally. So take the record

880

# size and memory available into consideration when setting this parameter.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

881

"maxWorkerCount": "A String", # Optional. The maximum number of workers to be used for parallel processing.

882

# Defaults to 10 if not specified.

883

},

884

"labels": { # Optional. One or more labels that you can add, to organize your jobs.

885

# Each label is a key-value pair, where both the key and the value are

886

# arbitrary strings that you supply.

887

# For more information, see the documentation on

888

# <a href="/ml-engine/docs/tensorflow/resource-labels">using labels</a>.

889

"a_key": "A String",

890

},

891

"endTime": "A String", # Output only. When the job processing was completed.

892

"trainingOutput": { # Represents results of a training job. Output only. # The current training job result.

893

"trials": [ # Results for individual Hyperparameter trials.

894

# Only set for hyperparameter tuning jobs.

895

{ # Represents the result of a single hyperparameter tuning trial from a

896

# training job. The TrainingOutput object that is returned on successful

897

# completion of a training job with hyperparameter tuning includes a list

898

# of HyperparameterOutput objects, one for each successful trial.

899

"endTime": "A String", # Output only. End time for the trial.

900

"finalMetric": { # An observed value of a metric. # The final objective metric seen for this trial.

901

"trainingStep": "A String", # The global training step for this metric.

902

"objectiveValue": 3.14, # The objective value at this training step.

903

},

904

"hyperparameters": { # The hyperparameters given to this trial.

905

"a_key": "A String",

906

},

907

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

908

# Only set for trials of built-in algorithms jobs that have succeeded.

909

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

910

# saves the trained model. Only set for successful jobs that don't use

911

# hyperparameter tuning.

912

"framework": "A String", # Framework on which the built-in algorithm was trained.

913

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

914

# trained.

915

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

916

},

917

"startTime": "A String", # Output only. Start time for the trial.

918

"allMetrics": [ # All recorded object metrics for this trial. This field is not currently

919

# populated.

920

{ # An observed value of a metric.

921

"trainingStep": "A String", # The global training step for this metric.

922

"objectiveValue": 3.14, # The objective value at this training step.

923

},

924

],

925

"trialId": "A String", # The trial id for these results.

926

"isTrialStoppedEarly": True or False, # True if the trial is stopped early.

927

"state": "A String", # Output only. The detailed state of the trial.

928

},

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

929

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

930

"completedTrialCount": "A String", # The number of hyperparameter tuning trials that completed successfully.

931

# Only set for hyperparameter tuning jobs.

932

"isHyperparameterTuningJob": True or False, # Whether this job is a hyperparameter tuning job.

933

"isBuiltInAlgorithmJob": True or False, # Whether this job is a built-in Algorithm job.

934

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

935

# Only set for built-in algorithms jobs.

936

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

937

# saves the trained model. Only set for successful jobs that don't use

938

# hyperparameter tuning.

939

"framework": "A String", # Framework on which the built-in algorithm was trained.

940

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

941

# trained.

942

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

943

},

944

"consumedMLUnits": 3.14, # The amount of ML units consumed by the job.

945

"hyperparameterMetricTag": "A String", # The TensorFlow summary tag name used for optimizing hyperparameter tuning

946

# trials. See

947

# [`HyperparameterSpec.hyperparameterMetricTag`](#HyperparameterSpec.FIELDS.hyperparameter_metric_tag)

948

# for more information. Only set for hyperparameter tuning jobs.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

949

},

950

"errorMessage": "A String", # Output only. The details of a failure or a cancellation.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

951

"predictionOutput": { # Represents results of a prediction job. # The current prediction job result.

952

"errorCount": "A String", # The number of data instances which resulted in errors.

953

"nodeHours": 3.14, # Node hours used by the batch prediction job.

954

"outputPath": "A String", # The output Google Cloud Storage location provided at the job creation time.

955

"predictionCount": "A String", # The number of generated predictions.

956

},

957

"startTime": "A String", # Output only. When the job processing was started.

958

"state": "A String", # Output only. The detailed state of a job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

959

"trainingInput": { # Represents input parameters for a training job. When using the gcloud command # Input parameters to create a training job.

960

# to submit your training job, you can specify the input parameters as

961

# command-line arguments and/or in a YAML configuration file referenced from

962

# the --config command-line argument. For details, see the guide to [submitting

963

# a training job](/ai-platform/training/docs/training-jobs).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

964

"serviceAccount": "A String", # Optional. Specifies the service account for workload run-as account.

965

# Users submitting jobs must have act-as permission on this run-as account.

966

# If not specified, then CMLE P4SA will be used by default.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

967

"workerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for workers.

968

#

969

# You should only set `workerConfig.acceleratorConfig` if `workerType` is set

970

# to a Compute Engine machine type. [Learn about restrictions on accelerator

971

# configurations for

972

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

973

#

974

# Set `workerConfig.imageUri` only if you build a custom image for your

975

# worker. If `workerConfig.imageUri` has not been set, AI Platform uses

976

# the value of `masterConfig.imageUri`. Learn more about [configuring custom

977

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

978

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

979

# [Learn about restrictions on accelerator configurations for

980

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

981

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

982

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

983

# [accelerators for online

984

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

985

"type": "A String", # The type of accelerator to use.

986

"count": "A String", # The number of accelerators to attach to each machine running the job.

987

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

988

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

989

# Registry. Learn more about [configuring custom

990

# containers](/ai-platform/training/docs/distributed-training-containers).

991

"containerArgs": [ # Arguments to the entrypoint command.

992

# The following rules apply for container_command and container_args:

993

# - If you do not supply command or args:

994

# The defaults defined in the Docker image are used.

995

# - If you supply a command but no args:

996

# The default EntryPoint and the default Cmd defined in the Docker image

997

# are ignored. Your command is run without any arguments.

998

# - If you supply only args:

999

# The default Entrypoint defined in the Docker image is run with the args

1000

# that you supplied.

1001

# - If you supply a command and args:

1002

# The default Entrypoint and the default Cmd defined in the Docker image

1003

# are ignored. Your command is run with your args.

1004

# It cannot be set if custom container image is

1005

# not provided.

1006

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1007

# both cannot be set at the same time.

1008

"A String",

1009

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1010

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

1011

# the one used in the custom container. This field is required if the replica

1012

# is a TPU worker that uses a custom container. Otherwise, do not specify

1013

# this field. This must be a [runtime version that currently supports

1014

# training with

1015

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

1016

#

1017

# Note that the version of TensorFlow included in a runtime version may

1018

# differ from the numbering of the runtime version itself, because it may

1019

# have a different [patch

1020

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

1021

# In this field, you must specify the runtime version (TensorFlow minor

1022

# version). For example, if your custom container runs TensorFlow `1.x.y`,

1023

# specify `1.x`.

1024

"containerCommand": [ # The command with which the replica's custom container is run.

1025

# If provided, it will override default ENTRYPOINT of the docker image.

1026

# If not provided, the docker image's ENTRYPOINT is used.

1027

# It cannot be set if custom container image is

1028

# not provided.

1029

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1030

# both cannot be set at the same time.

1031

"A String",

1032

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1033

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1034

"useChiefInTfConfig": True or False, # Optional. Use `chief` instead of `master` in the `TF_CONFIG` environment

1035

# variable when training with a custom container. Defaults to `false`. [Learn

1036

# more about this

1037

# field.](/ai-platform/training/docs/distributed-training-details#chief-versus-master)

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1038

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1039

# This field has no effect for training jobs that don't use a custom

1040

# container.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1041

"masterType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1042

# job's master worker. You must specify this field when `scaleTier` is set to

1043

# `CUSTOM`.

1044

#

1045

# You can use certain Compute Engine machine types directly in this field.

1046

# The following types are supported:

#

# - `n1-standard-4`

# - `n1-standard-8`

# - `n1-standard-16`

# - `n1-standard-32`

# - `n1-standard-64`

# - `n1-standard-96`

# - `n1-highmem-2`

# - `n1-highmem-4`

# - `n1-highmem-8`

# - `n1-highmem-16`

# - `n1-highmem-32`

# - `n1-highmem-64`

# - `n1-highmem-96`

# - `n1-highcpu-16`

# - `n1-highcpu-32`

# - `n1-highcpu-64`

# - `n1-highcpu-96`

#

# Learn more about [using Compute Engine machine

1067

# types](/ml-engine/docs/machine-types#compute-engine-machine-types).

1068

#

1069

# Alternatively, you can use the following legacy machine types:

#

# - `standard`

# - `large_model`

# - `complex_model_s`

1074

# - `complex_model_m`

1075

# - `complex_model_l`

1076

# - `standard_gpu`

1077

# - `complex_model_m_gpu`

1078

# - `complex_model_l_gpu`

1079

# - `standard_p100`

1080

# - `complex_model_m_p100`

1081

# - `standard_v100`

1082

# - `large_model_v100`

1083

# - `complex_model_m_v100`

1084

# - `complex_model_l_v100`

1085

#

1086

# Learn more about [using legacy machine

1087

# types](/ml-engine/docs/machine-types#legacy-machine-types).

1088

#

1089

# Finally, if you want to use a TPU for training, specify `cloud_tpu` in this

1090

# field. Learn more about the [special configuration options for training

1091

# with

1092

# TPUs](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1093

"parameterServerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for parameter servers.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1094

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1095

# You should only set `parameterServerConfig.acceleratorConfig` if

1096

# `parameterServerType` is set to a Compute Engine machine type. [Learn

1097

# about restrictions on accelerator configurations for

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1098

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1099

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1100

# Set `parameterServerConfig.imageUri` only if you build a custom image for

1101

# your parameter server. If `parameterServerConfig.imageUri` has not been

1102

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1103

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1104

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

1105

# [Learn about restrictions on accelerator configurations for

1106

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1107

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

1108

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

1109

# [accelerators for online

1110

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

1111

"type": "A String", # The type of accelerator to use.

1112

"count": "A String", # The number of accelerators to attach to each machine running the job.

1113

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1114

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

1115

# Registry. Learn more about [configuring custom

1116

# containers](/ai-platform/training/docs/distributed-training-containers).

1117

"containerArgs": [ # Arguments to the entrypoint command.

1118

# The following rules apply for container_command and container_args:

1119

# - If you do not supply command or args:

1120

# The defaults defined in the Docker image are used.

1121

# - If you supply a command but no args:

1122

# The default EntryPoint and the default Cmd defined in the Docker image

1123

# are ignored. Your command is run without any arguments.

1124

# - If you supply only args:

1125

# The default Entrypoint defined in the Docker image is run with the args

1126

# that you supplied.

1127

# - If you supply a command and args:

1128

# The default Entrypoint and the default Cmd defined in the Docker image

1129

# are ignored. Your command is run with your args.

1130

# It cannot be set if custom container image is

1131

# not provided.

1132

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1133

# both cannot be set at the same time.

1134

"A String",

1135

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1136

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

1137

# the one used in the custom container. This field is required if the replica

1138

# is a TPU worker that uses a custom container. Otherwise, do not specify

1139

# this field. This must be a [runtime version that currently supports

1140

# training with

1141

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

1142

#

1143

# Note that the version of TensorFlow included in a runtime version may

1144

# differ from the numbering of the runtime version itself, because it may

1145

# have a different [patch

1146

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

1147

# In this field, you must specify the runtime version (TensorFlow minor

1148

# version). For example, if your custom container runs TensorFlow `1.x.y`,

1149

# specify `1.x`.

1150

"containerCommand": [ # The command with which the replica's custom container is run.

1151

# If provided, it will override default ENTRYPOINT of the docker image.

1152

# If not provided, the docker image's ENTRYPOINT is used.

1153

# It cannot be set if custom container image is

1154

# not provided.

1155

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1156

# both cannot be set at the same time.

1157

"A String",

1158

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1159

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1160

"region": "A String", # Required. The region to run the training job in. See the [available

1161

# regions](/ai-platform/training/docs/regions) for AI Platform Training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1162

"jobDir": "A String", # Optional. A Google Cloud Storage path in which to store training outputs

1163

# and other data needed for training. This path is passed to your TensorFlow

1164

# program as the '--job-dir' command-line argument. The benefit of specifying

1165

# this field is that Cloud ML validates the path for use in training.

1166

"pythonVersion": "A String", # Optional. The version of Python used in training. You must either specify

1167

# this field or specify `masterConfig.imageUri`.

1168

#

1169

# The following Python versions are available:

1170

#

1171

# * Python '3.7' is available when `runtime_version` is set to '1.15' or

1172

# later.

1173

# * Python '3.5' is available when `runtime_version` is set to a version

1174

# from '1.4' to '1.14'.

1175

# * Python '2.7' is available when `runtime_version` is set to '1.15' or

1176

# earlier.

1177

#

1178

# Read more about the Python versions available for [each runtime

1179

# version](/ml-engine/docs/runtime-version-list).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1180

"hyperparameters": { # Represents a set of hyperparameters to optimize. # Optional. The set of Hyperparameters to tune.

1181

"maxTrials": 42, # Optional. How many training trials should be attempted to optimize

1182

# the specified hyperparameters.

1183

#

1184

# Defaults to one.

1185

"enableTrialEarlyStopping": True or False, # Optional. Indicates if the hyperparameter tuning job enables auto trial

1186

# early stopping.

1187

"params": [ # Required. The set of parameters to tune.

1188

{ # Represents a single hyperparameter to optimize.

1189

"minValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

1190

# should be unset if type is `CATEGORICAL`. This value should be integers if

1191

# type is INTEGER.

1192

"categoricalValues": [ # Required if type is `CATEGORICAL`. The list of possible categories.

1193

"A String",

1194

],

1195

"scaleType": "A String", # Optional. How the parameter should be scaled to the hypercube.

1196

# Leave unset for categorical parameters.

1197

# Some kind of scaling is strongly recommended for real or integral

1198

# parameters (e.g., `UNIT_LINEAR_SCALE`).

1199

"discreteValues": [ # Required if type is `DISCRETE`.

1200

# A list of feasible points.

1201

# The list should be in strictly increasing order. For instance, this

1202

# parameter might have possible settings of 1.5, 2.5, and 4.0. This list

1203

# should not contain more than 1,000 values.

1204

3.14,

1205

],

1206

"type": "A String", # Required. The type of the parameter.

1207

"maxValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

1208

# should be unset if type is `CATEGORICAL`. This value should be integers if

1209

# type is `INTEGER`.

1210

"parameterName": "A String", # Required. The parameter name must be unique amongst all ParameterConfigs in

1211

# a HyperparameterSpec message. E.g., "learning_rate".

1212

},

1213

],

1214

"maxFailedTrials": 42, # Optional. The number of failed trials that need to be seen before failing

1215

# the hyperparameter tuning job. You can specify this field to override the

1216

# default failing criteria for AI Platform hyperparameter tuning jobs.

1217

#

1218

# Defaults to zero, which means the service decides when a hyperparameter

1219

# job should fail.

1220

"hyperparameterMetricTag": "A String", # Optional. The TensorFlow summary tag name to use for optimizing trials. For

1221

# current versions of TensorFlow, this tag name should exactly match what is

1222

# shown in TensorBoard, including all scopes. For versions of TensorFlow

1223

# prior to 0.12, this should be only the tag passed to tf.Summary.

1224

# By default, "training/hptuning/metric" will be used.

1225

"resumePreviousJobId": "A String", # Optional. The prior hyperparameter tuning job id that users hope to

1226

# continue with. The job id will be used to find the corresponding vizier

1227

# study guid and resume the study.

1228

"goal": "A String", # Required. The type of goal to use for tuning. Available types are

1229

# `MAXIMIZE` and `MINIMIZE`.

1230

#

1231

# Defaults to `MAXIMIZE`.

1232

"algorithm": "A String", # Optional. The search algorithm specified for the hyperparameter

1233

# tuning job.

1234

# Uses the default AI Platform hyperparameter tuning

1235

# algorithm if unspecified.

1236

"maxParallelTrials": 42, # Optional. The number of training trials to run concurrently.

1237

# You can reduce the time it takes to perform hyperparameter tuning by adding

1238

# trials in parallel. However, each trail only benefits from the information

1239

# gained in completed trials. That means that a trial does not get access to

1240

# the results of trials running at the same time, which could reduce the

1241

# quality of the overall optimization.

1242

#

1243

# Each trial will use the same scale tier and machine types.

#

# Defaults to one.

},

"evaluatorType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1248

# job's evaluator nodes.

1249

#

1250

# The supported values are the same as those described in the entry for

1251

# `masterType`.

1252

#

1253

# This value must be consistent with the category of machine type that

1254

# `masterType` uses. In other words, both must be Compute Engine machine

1255

# types or both must be legacy machine types.

1256

#

1257

# This value must be present when `scaleTier` is set to `CUSTOM` and

1258

# `evaluatorCount` is greater than zero.

1259

"network": "A String", # Optional. The full name of the Google Compute Engine

1260

# [network](/compute/docs/networks-and-firewalls#networks) to which the Job

1261

# is peered. For example, projects/12345/global/networks/myVPC. Format is of

1262

# the form projects/{project}/global/networks/{network}. Where {project} is a

1263

# project number, as in '12345', and {network} is network name.".

1264

#

1265

# Private services access must already be configured for the network. If left

1266

# unspecified, the Job is not peered with any network. Learn more -

1267

# Connecting Job to user network over private

1268

# IP.

1269

"parameterServerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1270

# job's parameter server.

1271

#

1272

# The supported values are the same as those described in the entry for

1273

# `master_type`.

1274

#

1275

# This value must be consistent with the category of machine type that

1276

# `masterType` uses. In other words, both must be Compute Engine machine

1277

# types or both must be legacy machine types.

1278

#

1279

# This value must be present when `scaleTier` is set to `CUSTOM` and

1280

# `parameter_server_count` is greater than zero.

1281

"workerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1282

# job's worker nodes.

1283

#

1284

# The supported values are the same as those described in the entry for

1285

# `masterType`.

1286

#

1287

# This value must be consistent with the category of machine type that

1288

# `masterType` uses. In other words, both must be Compute Engine machine

1289

# types or both must be legacy machine types.

1290

#

1291

# If you use `cloud_tpu` for this value, see special instructions for

1292

# [configuring a custom TPU

1293

# machine](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

1294

#

1295

# This value must be present when `scaleTier` is set to `CUSTOM` and

1296

# `workerCount` is greater than zero.

1297

"masterConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for your master worker.

1298

#

1299

# You should only set `masterConfig.acceleratorConfig` if `masterType` is set

1300

# to a Compute Engine machine type. Learn about [restrictions on accelerator

1301

# configurations for

1302

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1303

#

1304

# Set `masterConfig.imageUri` only if you build a custom image. Only one of

1305

# `masterConfig.imageUri` and `runtimeVersion` should be set. Learn more

1306

# about [configuring custom

1307

# containers](/ai-platform/training/docs/distributed-training-containers).

1308

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

1309

# [Learn about restrictions on accelerator configurations for

1310

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1311

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

1312

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

1313

# [accelerators for online

1314

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

1315

"type": "A String", # The type of accelerator to use.

1316

"count": "A String", # The number of accelerators to attach to each machine running the job.

1317

},

1318

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

1319

# Registry. Learn more about [configuring custom

1320

# containers](/ai-platform/training/docs/distributed-training-containers).

1321

"containerArgs": [ # Arguments to the entrypoint command.

1322

# The following rules apply for container_command and container_args:

1323

# - If you do not supply command or args:

1324

# The defaults defined in the Docker image are used.

1325

# - If you supply a command but no args:

1326

# The default EntryPoint and the default Cmd defined in the Docker image

1327

# are ignored. Your command is run without any arguments.

1328

# - If you supply only args:

1329

# The default Entrypoint defined in the Docker image is run with the args

1330

# that you supplied.

1331

# - If you supply a command and args:

1332

# The default Entrypoint and the default Cmd defined in the Docker image

1333

# are ignored. Your command is run with your args.

1334

# It cannot be set if custom container image is

1335

# not provided.

1336

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1337

# both cannot be set at the same time.

1338

"A String",

1339

],

1340

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

1341

# the one used in the custom container. This field is required if the replica

1342

# is a TPU worker that uses a custom container. Otherwise, do not specify

1343

# this field. This must be a [runtime version that currently supports

1344

# training with

1345

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

1346

#

1347

# Note that the version of TensorFlow included in a runtime version may

1348

# differ from the numbering of the runtime version itself, because it may

1349

# have a different [patch

1350

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

1351

# In this field, you must specify the runtime version (TensorFlow minor

1352

# version). For example, if your custom container runs TensorFlow `1.x.y`,

1353

# specify `1.x`.

1354

"containerCommand": [ # The command with which the replica's custom container is run.

1355

# If provided, it will override default ENTRYPOINT of the docker image.

1356

# If not provided, the docker image's ENTRYPOINT is used.

1357

# It cannot be set if custom container image is

1358

# not provided.

1359

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1360

# both cannot be set at the same time.

"A String",

],

},

"evaluatorCount": "A String", # Optional. The number of evaluator replicas to use for the training job.

1365

# Each replica in the cluster will be of the type specified in

1366

# `evaluator_type`.

1367

#

1368

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

1369

# set this value, you must also set `evaluator_type`.

1370

#

1371

# The default value is zero.

1372

"args": [ # Optional. Command-line arguments passed to the training application when it

1373

# starts. If your job uses a custom container, then the arguments are passed

1374

# to the container's <a class="external" target="_blank"

1375

# href="https://docs.docker.com/engine/reference/builder/#entrypoint">

1376

# `ENTRYPOINT`</a> command.

1377

"A String",

1378

],

1379

"pythonModule": "A String", # Required. The Python module name to run after installing the packages.

1380

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for training. You must

1381

# either specify this field or specify `masterConfig.imageUri`.

1382

#

1383

# For more information, see the [runtime version

1384

# list](/ai-platform/training/docs/runtime-version-list) and learn [how to

1385

# manage runtime versions](/ai-platform/training/docs/versioning).

1386

"parameterServerCount": "A String", # Optional. The number of parameter server replicas to use for the training

1387

# job. Each replica in the cluster will be of the type specified in

1388

# `parameter_server_type`.

1389

#

1390

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

1391

# set this value, you must also set `parameter_server_type`.

1392

#

1393

# The default value is zero.

1394

"evaluatorConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for evaluators.

1395

#

1396

# You should only set `evaluatorConfig.acceleratorConfig` if

1397

# `evaluatorType` is set to a Compute Engine machine type. [Learn

1398

# about restrictions on accelerator configurations for

1399

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1400

#

1401

# Set `evaluatorConfig.imageUri` only if you build a custom image for

1402

# your evaluator. If `evaluatorConfig.imageUri` has not been

1403

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

1404

# containers](/ai-platform/training/docs/distributed-training-containers).

1405

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

1406

# [Learn about restrictions on accelerator configurations for

1407

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1408

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

1409

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

1410

# [accelerators for online

1411

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

1412

"type": "A String", # The type of accelerator to use.

1413

"count": "A String", # The number of accelerators to attach to each machine running the job.

1414

},

1415

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

1416

# Registry. Learn more about [configuring custom

1417

# containers](/ai-platform/training/docs/distributed-training-containers).

1418

"containerArgs": [ # Arguments to the entrypoint command.

1419

# The following rules apply for container_command and container_args:

1420

# - If you do not supply command or args:

1421

# The defaults defined in the Docker image are used.

1422

# - If you supply a command but no args:

1423

# The default EntryPoint and the default Cmd defined in the Docker image

1424

# are ignored. Your command is run without any arguments.

1425

# - If you supply only args:

1426

# The default Entrypoint defined in the Docker image is run with the args

1427

# that you supplied.

1428

# - If you supply a command and args:

1429

# The default Entrypoint and the default Cmd defined in the Docker image

1430

# are ignored. Your command is run with your args.

1431

# It cannot be set if custom container image is

1432

# not provided.

1433

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1434

# both cannot be set at the same time.

1435

"A String",

1436

],

1437

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

1438

# the one used in the custom container. This field is required if the replica

1439

# is a TPU worker that uses a custom container. Otherwise, do not specify

1440

# this field. This must be a [runtime version that currently supports

1441

# training with

1442

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

1443

#

1444

# Note that the version of TensorFlow included in a runtime version may

1445

# differ from the numbering of the runtime version itself, because it may

1446

# have a different [patch

1447

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

1448

# In this field, you must specify the runtime version (TensorFlow minor

1449

# version). For example, if your custom container runs TensorFlow `1.x.y`,

1450

# specify `1.x`.

1451

"containerCommand": [ # The command with which the replica's custom container is run.

1452

# If provided, it will override default ENTRYPOINT of the docker image.

1453

# If not provided, the docker image's ENTRYPOINT is used.

1454

# It cannot be set if custom container image is

1455

# not provided.

1456

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1457

# both cannot be set at the same time.

"A String",

],

},

"encryptionConfig": { # Represents a custom encryption key configuration that can be applied to # Optional. Options for using customer-managed encryption keys (CMEK) to

1462

# protect resources created by a training job, instead of using Google's

1463

# default encryption. If this is set, then all resources created by the

1464

# training job will be encrypted with the customer-managed encryption key

1465

# that you specify.

1466

#

1467

# [Learn how and when to use CMEK with AI Platform

1468

# Training](/ai-platform/training/docs/cmek).

1469

# a resource.

1470

"kmsKeyName": "A String", # The Cloud KMS resource identifier of the customer-managed encryption key

1471

# used to protect a resource, such as a training job. It has the following

1472

# format:

1473

# `projects/{PROJECT_ID}/locations/{REGION}/keyRings/{KEY_RING_NAME}/cryptoKeys/{KEY_NAME}`

1474

},

1475

"workerCount": "A String", # Optional. The number of worker replicas to use for the training job. Each

1476

# replica in the cluster will be of the type specified in `worker_type`.

1477

#

1478

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

1479

# set this value, you must also set `worker_type`.

1480

#

1481

# The default value is zero.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1482

"scheduling": { # All parameters related to scheduling of training jobs. # Optional. Scheduling options for a training job.

1483

"maxWaitTime": "A String",

1484

"maxRunningTime": "A String", # Optional. The maximum job running time, expressed in seconds. The field can

1485

# contain up to nine fractional digits, terminated by `s`. If not specified,

1486

# this field defaults to `604800s` (seven days).

1487

#

1488

# If the training job is still running after this duration, AI Platform

1489

# Training cancels it.

1490

#

1491

# For example, if you want to ensure your job runs for no more than 2 hours,

1492

# set this field to `7200s` (2 hours * 60 minutes / hour * 60 seconds /

1493

# minute).

1494

#

1495

# If you submit your training job using the `gcloud` tool, you can [provide

1496

# this field in a `config.yaml`

1497

# file](/ai-platform/training/docs/training-jobs#formatting_your_configuration_parameters).

# For example:

#

# ```yaml

# trainingInput:

# ...

# scheduling:

# maxRunningTime: 7200s

1505

# ...

1506

# ```

1507

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1508

"scaleTier": "A String", # Required. Specifies the machine types, the number of replicas for workers

1509

# and parameter servers.

1510

"packageUris": [ # Required. The Google Cloud Storage location of the packages with

1511

# the training program and any additional dependencies.

1512

# The maximum number of package URIs is 100.

1513

"A String",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1514

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1515

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1516

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

1517

# prevent simultaneous updates of a job from overwriting each other.

1518

# It is strongly suggested that systems make use of the `etag` in the

1519

# read-modify-write cycle to perform job updates in order to avoid race

1520

# conditions: An `etag` is returned in the response to `GetJob`, and

1521

# systems are expected to put that etag in the request to `UpdateJob` to

1522

# ensure that their change will be applied to the same version of the job.

1523

"jobId": "A String", # Required. The user-specified id of the job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1524

}</pre>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

</div>

<code class="details" id="get">get(name, x__xgafv=None)</code>

1529

<pre>Describes a job.

1530

1531

Args:

1532

name: string, Required. The name of the job to get the description of. (required)

1533

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

1540

1541

{ # Represents a training or prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1542

"createTime": "A String", # Output only. When the job was created.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1543

"predictionInput": { # Represents input parameters for a prediction job. # Input parameters to create a prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1544

"dataFormat": "A String", # Required. The format of the input data files.

1545

"outputPath": "A String", # Required. The output Google Cloud Storage location.

1546

"inputPaths": [ # Required. The Cloud Storage location of the input data files. May contain

1547

# <a href="/storage/docs/gsutil/addlhelp/WildcardNames">wildcards</a>.

1548

"A String",

1549

],

1550

"region": "A String", # Required. The Google Compute Engine region to run the prediction job in.

1551

# See the <a href="/ml-engine/docs/tensorflow/regions">available regions</a>

1552

# for AI Platform services.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1553

"versionName": "A String", # Use this field if you want to specify a version of the model to use. The

1554

# string is formatted the same way as `model_version`, with the addition

1555

# of the version information:

1556

#

1557

# `"projects/YOUR_PROJECT/models/YOUR_MODEL/versions/YOUR_VERSION"`

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1558

"uri": "A String", # Use this field if you want to specify a Google Cloud Storage path for

1559

# the model to use.

1560

"outputDataFormat": "A String", # Optional. Format of the output data files, defaults to JSON.

1561

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for this batch

1562

# prediction. If not set, AI Platform will pick the runtime version used

1563

# during the CreateVersion request for this model version, or choose the

1564

# latest stable version when model version information is not available

1565

# such as when the model is specified by uri.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1566

"modelName": "A String", # Use this field if you want to use the default version for the specified

1567

# model. The string must use the following format:

1568

#

1569

# `"projects/YOUR_PROJECT/models/YOUR_MODEL"`

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1570

"signatureName": "A String", # Optional. The name of the signature defined in the SavedModel to use for

1571

# this job. Please refer to

1572

# [SavedModel](https://tensorflow.github.io/serving/serving_basic.html)

1573

# for information about how to use signatures.

1574

#

1575

# Defaults to

1576

# [DEFAULT_SERVING_SIGNATURE_DEF_KEY](https://www.tensorflow.org/api_docs/python/tf/saved_model/signature_constants)

1577

# , which is "serving_default".

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1578

"batchSize": "A String", # Optional. Number of records per batch, defaults to 64.

1579

# The service will buffer batch_size number of records in memory before

1580

# invoking one Tensorflow prediction call internally. So take the record

1581

# size and memory available into consideration when setting this parameter.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1582

"maxWorkerCount": "A String", # Optional. The maximum number of workers to be used for parallel processing.

1583

# Defaults to 10 if not specified.

1584

},

1585

"labels": { # Optional. One or more labels that you can add, to organize your jobs.

1586

# Each label is a key-value pair, where both the key and the value are

1587

# arbitrary strings that you supply.

1588

# For more information, see the documentation on

1589

# <a href="/ml-engine/docs/tensorflow/resource-labels">using labels</a>.

1590

"a_key": "A String",

1591

},

1592

"endTime": "A String", # Output only. When the job processing was completed.

1593

"trainingOutput": { # Represents results of a training job. Output only. # The current training job result.

1594

"trials": [ # Results for individual Hyperparameter trials.

1595

# Only set for hyperparameter tuning jobs.

1596

{ # Represents the result of a single hyperparameter tuning trial from a

1597

# training job. The TrainingOutput object that is returned on successful

1598

# completion of a training job with hyperparameter tuning includes a list

1599

# of HyperparameterOutput objects, one for each successful trial.

1600

"endTime": "A String", # Output only. End time for the trial.

1601

"finalMetric": { # An observed value of a metric. # The final objective metric seen for this trial.

1602

"trainingStep": "A String", # The global training step for this metric.

1603

"objectiveValue": 3.14, # The objective value at this training step.

1604

},

1605

"hyperparameters": { # The hyperparameters given to this trial.

1606

"a_key": "A String",

1607

},

1608

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

1609

# Only set for trials of built-in algorithms jobs that have succeeded.

1610

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

1611

# saves the trained model. Only set for successful jobs that don't use

1612

# hyperparameter tuning.

1613

"framework": "A String", # Framework on which the built-in algorithm was trained.

1614

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

1615

# trained.

1616

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

1617

},

1618

"startTime": "A String", # Output only. Start time for the trial.

1619

"allMetrics": [ # All recorded object metrics for this trial. This field is not currently

1620

# populated.

1621

{ # An observed value of a metric.

1622

"trainingStep": "A String", # The global training step for this metric.

1623

"objectiveValue": 3.14, # The objective value at this training step.

1624

},

1625

],

1626

"trialId": "A String", # The trial id for these results.

1627

"isTrialStoppedEarly": True or False, # True if the trial is stopped early.

1628

"state": "A String", # Output only. The detailed state of the trial.

1629

},

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1630

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1631

"completedTrialCount": "A String", # The number of hyperparameter tuning trials that completed successfully.

1632

# Only set for hyperparameter tuning jobs.

1633

"isHyperparameterTuningJob": True or False, # Whether this job is a hyperparameter tuning job.

1634

"isBuiltInAlgorithmJob": True or False, # Whether this job is a built-in Algorithm job.

1635

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

1636

# Only set for built-in algorithms jobs.

1637

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

1638

# saves the trained model. Only set for successful jobs that don't use

1639

# hyperparameter tuning.

1640

"framework": "A String", # Framework on which the built-in algorithm was trained.

1641

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

1642

# trained.

1643

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

1644

},

1645

"consumedMLUnits": 3.14, # The amount of ML units consumed by the job.

1646

"hyperparameterMetricTag": "A String", # The TensorFlow summary tag name used for optimizing hyperparameter tuning

1647

# trials. See

1648

# [`HyperparameterSpec.hyperparameterMetricTag`](#HyperparameterSpec.FIELDS.hyperparameter_metric_tag)

1649

# for more information. Only set for hyperparameter tuning jobs.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1650

},

1651

"errorMessage": "A String", # Output only. The details of a failure or a cancellation.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1652

"predictionOutput": { # Represents results of a prediction job. # The current prediction job result.

1653

"errorCount": "A String", # The number of data instances which resulted in errors.

1654

"nodeHours": 3.14, # Node hours used by the batch prediction job.

1655

"outputPath": "A String", # The output Google Cloud Storage location provided at the job creation time.

1656

"predictionCount": "A String", # The number of generated predictions.

1657

},

1658

"startTime": "A String", # Output only. When the job processing was started.

1659

"state": "A String", # Output only. The detailed state of a job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1660

"trainingInput": { # Represents input parameters for a training job. When using the gcloud command # Input parameters to create a training job.

1661

# to submit your training job, you can specify the input parameters as

1662

# command-line arguments and/or in a YAML configuration file referenced from

1663

# the --config command-line argument. For details, see the guide to [submitting

1664

# a training job](/ai-platform/training/docs/training-jobs).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1665

"serviceAccount": "A String", # Optional. Specifies the service account for workload run-as account.

1666

# Users submitting jobs must have act-as permission on this run-as account.

1667

# If not specified, then CMLE P4SA will be used by default.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1668

"workerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for workers.

1669

#

1670

# You should only set `workerConfig.acceleratorConfig` if `workerType` is set

1671

# to a Compute Engine machine type. [Learn about restrictions on accelerator

1672

# configurations for

1673

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1674

#

1675

# Set `workerConfig.imageUri` only if you build a custom image for your

1676

# worker. If `workerConfig.imageUri` has not been set, AI Platform uses

1677

# the value of `masterConfig.imageUri`. Learn more about [configuring custom

1678

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1679

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

1680

# [Learn about restrictions on accelerator configurations for

1681

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1682

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

1683

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

1684

# [accelerators for online

1685

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

1686

"type": "A String", # The type of accelerator to use.

1687

"count": "A String", # The number of accelerators to attach to each machine running the job.

1688

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1689

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

1690

# Registry. Learn more about [configuring custom

1691

# containers](/ai-platform/training/docs/distributed-training-containers).

1692

"containerArgs": [ # Arguments to the entrypoint command.

1693

# The following rules apply for container_command and container_args:

1694

# - If you do not supply command or args:

1695

# The defaults defined in the Docker image are used.

1696

# - If you supply a command but no args:

1697

# The default EntryPoint and the default Cmd defined in the Docker image

1698

# are ignored. Your command is run without any arguments.

1699

# - If you supply only args:

1700

# The default Entrypoint defined in the Docker image is run with the args

1701

# that you supplied.

1702

# - If you supply a command and args:

1703

# The default Entrypoint and the default Cmd defined in the Docker image

1704

# are ignored. Your command is run with your args.

1705

# It cannot be set if custom container image is

1706

# not provided.

1707

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1708

# both cannot be set at the same time.

1709

"A String",

1710

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1711

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

1712

# the one used in the custom container. This field is required if the replica

1713

# is a TPU worker that uses a custom container. Otherwise, do not specify

1714

# this field. This must be a [runtime version that currently supports

1715

# training with

1716

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

1717

#

1718

# Note that the version of TensorFlow included in a runtime version may

1719

# differ from the numbering of the runtime version itself, because it may

1720

# have a different [patch

1721

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

1722

# In this field, you must specify the runtime version (TensorFlow minor

1723

# version). For example, if your custom container runs TensorFlow `1.x.y`,

1724

# specify `1.x`.

1725

"containerCommand": [ # The command with which the replica's custom container is run.

1726

# If provided, it will override default ENTRYPOINT of the docker image.

1727

# If not provided, the docker image's ENTRYPOINT is used.

1728

# It cannot be set if custom container image is

1729

# not provided.

1730

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1731

# both cannot be set at the same time.

1732

"A String",

1733

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

1734

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1735

"useChiefInTfConfig": True or False, # Optional. Use `chief` instead of `master` in the `TF_CONFIG` environment

1736

# variable when training with a custom container. Defaults to `false`. [Learn

1737

# more about this

1738

# field.](/ai-platform/training/docs/distributed-training-details#chief-versus-master)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

1739

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1740

# This field has no effect for training jobs that don't use a custom

1741

# container.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1742

"masterType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1743

# job's master worker. You must specify this field when `scaleTier` is set to

1744

# `CUSTOM`.

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

1745

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1746

# You can use certain Compute Engine machine types directly in this field.

1747

# The following types are supported:

#

# - `n1-standard-4`

# - `n1-standard-8`

# - `n1-standard-16`

# - `n1-standard-32`

# - `n1-standard-64`

# - `n1-standard-96`

# - `n1-highmem-2`

# - `n1-highmem-4`

# - `n1-highmem-8`

# - `n1-highmem-16`

# - `n1-highmem-32`

# - `n1-highmem-64`

# - `n1-highmem-96`

# - `n1-highcpu-16`

# - `n1-highcpu-32`

# - `n1-highcpu-64`

# - `n1-highcpu-96`

#

# Learn more about [using Compute Engine machine

1768

# types](/ml-engine/docs/machine-types#compute-engine-machine-types).

1769

#

1770

# Alternatively, you can use the following legacy machine types:

#

# - `standard`

# - `large_model`

# - `complex_model_s`

1775

# - `complex_model_m`

1776

# - `complex_model_l`

1777

# - `standard_gpu`

1778

# - `complex_model_m_gpu`

1779

# - `complex_model_l_gpu`

1780

# - `standard_p100`

1781

# - `complex_model_m_p100`

1782

# - `standard_v100`

1783

# - `large_model_v100`

1784

# - `complex_model_m_v100`

1785

# - `complex_model_l_v100`

1786

#

1787

# Learn more about [using legacy machine

1788

# types](/ml-engine/docs/machine-types#legacy-machine-types).

1789

#

1790

# Finally, if you want to use a TPU for training, specify `cloud_tpu` in this

1791

# field. Learn more about the [special configuration options for training

1792

# with

1793

# TPUs](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1794

"parameterServerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for parameter servers.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1795

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1796

# You should only set `parameterServerConfig.acceleratorConfig` if

1797

# `parameterServerType` is set to a Compute Engine machine type. [Learn

1798

# about restrictions on accelerator configurations for

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

1799

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

1800

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1801

# Set `parameterServerConfig.imageUri` only if you build a custom image for

1802

# your parameter server. If `parameterServerConfig.imageUri` has not been

1803

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1804

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1805

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

1806

# [Learn about restrictions on accelerator configurations for

1807

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

1808

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

1809

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

1810

# [accelerators for online

1811

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

1812

"type": "A String", # The type of accelerator to use.

1813

"count": "A String", # The number of accelerators to attach to each machine running the job.

1814

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1815

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

1816

# Registry. Learn more about [configuring custom

1817

# containers](/ai-platform/training/docs/distributed-training-containers).

1818

"containerArgs": [ # Arguments to the entrypoint command.

1819

# The following rules apply for container_command and container_args:

1820

# - If you do not supply command or args:

1821

# The defaults defined in the Docker image are used.

1822

# - If you supply a command but no args:

1823

# The default EntryPoint and the default Cmd defined in the Docker image

1824

# are ignored. Your command is run without any arguments.

1825

# - If you supply only args:

1826

# The default Entrypoint defined in the Docker image is run with the args

1827

# that you supplied.

1828

# - If you supply a command and args:

1829

# The default Entrypoint and the default Cmd defined in the Docker image

1830

# are ignored. Your command is run with your args.

1831

# It cannot be set if custom container image is

1832

# not provided.

1833

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1834

# both cannot be set at the same time.

1835

"A String",

1836

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

1837

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

1838

# the one used in the custom container. This field is required if the replica

1839

# is a TPU worker that uses a custom container. Otherwise, do not specify

1840

# this field. This must be a [runtime version that currently supports

1841

# training with

1842

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

1843

#

1844

# Note that the version of TensorFlow included in a runtime version may

1845

# differ from the numbering of the runtime version itself, because it may

1846

# have a different [patch

1847

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

1848

# In this field, you must specify the runtime version (TensorFlow minor

1849

# version). For example, if your custom container runs TensorFlow `1.x.y`,

1850

# specify `1.x`.

1851

"containerCommand": [ # The command with which the replica's custom container is run.

1852

# If provided, it will override default ENTRYPOINT of the docker image.

1853

# If not provided, the docker image's ENTRYPOINT is used.

1854

# It cannot be set if custom container image is

1855

# not provided.

1856

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

1857

# both cannot be set at the same time.

1858

"A String",

1859

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1860

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1861

"region": "A String", # Required. The region to run the training job in. See the [available

1862

# regions](/ai-platform/training/docs/regions) for AI Platform Training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

1863

"jobDir": "A String", # Optional. A Google Cloud Storage path in which to store training outputs

1864

# and other data needed for training. This path is passed to your TensorFlow

1865

# program as the '--job-dir' command-line argument. The benefit of specifying

1866

# this field is that Cloud ML validates the path for use in training.

1867

"pythonVersion": "A String", # Optional. The version of Python used in training. You must either specify

1868

# this field or specify `masterConfig.imageUri`.

1869

#

1870

# The following Python versions are available:

1871

#

1872

# * Python '3.7' is available when `runtime_version` is set to '1.15' or

1873

# later.

1874

# * Python '3.5' is available when `runtime_version` is set to a version

1875

# from '1.4' to '1.14'.

1876

# * Python '2.7' is available when `runtime_version` is set to '1.15' or

1877

# earlier.

1878

#

1879

# Read more about the Python versions available for [each runtime

1880

# version](/ml-engine/docs/runtime-version-list).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

1881

"hyperparameters": { # Represents a set of hyperparameters to optimize. # Optional. The set of Hyperparameters to tune.

1882

"maxTrials": 42, # Optional. How many training trials should be attempted to optimize

1883

# the specified hyperparameters.

1884

#

1885

# Defaults to one.

1886

"enableTrialEarlyStopping": True or False, # Optional. Indicates if the hyperparameter tuning job enables auto trial

1887

# early stopping.

1888

"params": [ # Required. The set of parameters to tune.

1889

{ # Represents a single hyperparameter to optimize.

1890

"minValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

1891

# should be unset if type is `CATEGORICAL`. This value should be integers if

1892

# type is INTEGER.

1893

"categoricalValues": [ # Required if type is `CATEGORICAL`. The list of possible categories.

1894

"A String",

1895

],

1896

"scaleType": "A String", # Optional. How the parameter should be scaled to the hypercube.

1897

# Leave unset for categorical parameters.

1898

# Some kind of scaling is strongly recommended for real or integral

1899

# parameters (e.g., `UNIT_LINEAR_SCALE`).

1900

"discreteValues": [ # Required if type is `DISCRETE`.

1901

# A list of feasible points.

1902

# The list should be in strictly increasing order. For instance, this

1903

# parameter might have possible settings of 1.5, 2.5, and 4.0. This list

1904

# should not contain more than 1,000 values.

1905

3.14,

1906

],

1907

"type": "A String", # Required. The type of the parameter.

1908

"maxValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

1909

# should be unset if type is `CATEGORICAL`. This value should be integers if

1910

# type is `INTEGER`.

1911

"parameterName": "A String", # Required. The parameter name must be unique amongst all ParameterConfigs in

1912

# a HyperparameterSpec message. E.g., "learning_rate".

1913

},

1914

],

1915

"maxFailedTrials": 42, # Optional. The number of failed trials that need to be seen before failing

1916

# the hyperparameter tuning job. You can specify this field to override the

1917

# default failing criteria for AI Platform hyperparameter tuning jobs.

1918

#

1919

# Defaults to zero, which means the service decides when a hyperparameter

1920

# job should fail.

1921

"hyperparameterMetricTag": "A String", # Optional. The TensorFlow summary tag name to use for optimizing trials. For

1922

# current versions of TensorFlow, this tag name should exactly match what is

1923

# shown in TensorBoard, including all scopes. For versions of TensorFlow

1924

# prior to 0.12, this should be only the tag passed to tf.Summary.

1925

# By default, "training/hptuning/metric" will be used.

1926

"resumePreviousJobId": "A String", # Optional. The prior hyperparameter tuning job id that users hope to

1927

# continue with. The job id will be used to find the corresponding vizier

1928

# study guid and resume the study.

1929

"goal": "A String", # Required. The type of goal to use for tuning. Available types are

1930

# `MAXIMIZE` and `MINIMIZE`.

1931

#

1932

# Defaults to `MAXIMIZE`.

1933

"algorithm": "A String", # Optional. The search algorithm specified for the hyperparameter

1934

# tuning job.

1935

# Uses the default AI Platform hyperparameter tuning

1936

# algorithm if unspecified.

1937

"maxParallelTrials": 42, # Optional. The number of training trials to run concurrently.

1938

# You can reduce the time it takes to perform hyperparameter tuning by adding

1939

# trials in parallel. However, each trail only benefits from the information

1940

# gained in completed trials. That means that a trial does not get access to

1941

# the results of trials running at the same time, which could reduce the

1942

# quality of the overall optimization.

1943

#

1944

# Each trial will use the same scale tier and machine types.

#

# Defaults to one.

},

"evaluatorType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1949

# job's evaluator nodes.

1950

#

1951

# The supported values are the same as those described in the entry for

1952

# `masterType`.

1953

#

1954

# This value must be consistent with the category of machine type that

1955

# `masterType` uses. In other words, both must be Compute Engine machine

1956

# types or both must be legacy machine types.

1957

#

1958

# This value must be present when `scaleTier` is set to `CUSTOM` and

1959

# `evaluatorCount` is greater than zero.

1960

"network": "A String", # Optional. The full name of the Google Compute Engine

1961

# [network](/compute/docs/networks-and-firewalls#networks) to which the Job

1962

# is peered. For example, projects/12345/global/networks/myVPC. Format is of

1963

# the form projects/{project}/global/networks/{network}. Where {project} is a

1964

# project number, as in '12345', and {network} is network name.".

1965

#

1966

# Private services access must already be configured for the network. If left

1967

# unspecified, the Job is not peered with any network. Learn more -

1968

# Connecting Job to user network over private

1969

# IP.

1970

"parameterServerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1971

# job's parameter server.

1972

#

1973

# The supported values are the same as those described in the entry for

1974

# `master_type`.

1975

#

1976

# This value must be consistent with the category of machine type that

1977

# `masterType` uses. In other words, both must be Compute Engine machine

1978

# types or both must be legacy machine types.

1979

#

1980

# This value must be present when `scaleTier` is set to `CUSTOM` and

1981

# `parameter_server_count` is greater than zero.

1982

"workerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

1983

# job's worker nodes.

1984

#

1985

# The supported values are the same as those described in the entry for

1986

# `masterType`.

1987

#

1988

# This value must be consistent with the category of machine type that

1989

# `masterType` uses. In other words, both must be Compute Engine machine

1990

# types or both must be legacy machine types.

1991

#

1992

# If you use `cloud_tpu` for this value, see special instructions for

1993

# [configuring a custom TPU

1994

# machine](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

1995

#

1996

# This value must be present when `scaleTier` is set to `CUSTOM` and

1997

# `workerCount` is greater than zero.

1998

"masterConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for your master worker.

1999

#

2000

# You should only set `masterConfig.acceleratorConfig` if `masterType` is set

2001

# to a Compute Engine machine type. Learn about [restrictions on accelerator

2002

# configurations for

2003

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

2004

#

2005

# Set `masterConfig.imageUri` only if you build a custom image. Only one of

2006

# `masterConfig.imageUri` and `runtimeVersion` should be set. Learn more

2007

# about [configuring custom

2008

# containers](/ai-platform/training/docs/distributed-training-containers).

2009

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

2010

# [Learn about restrictions on accelerator configurations for

2011

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

2012

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

2013

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

2014

# [accelerators for online

2015

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

2016

"type": "A String", # The type of accelerator to use.

2017

"count": "A String", # The number of accelerators to attach to each machine running the job.

2018

},

2019

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

2020

# Registry. Learn more about [configuring custom

2021

# containers](/ai-platform/training/docs/distributed-training-containers).

2022

"containerArgs": [ # Arguments to the entrypoint command.

2023

# The following rules apply for container_command and container_args:

2024

# - If you do not supply command or args:

2025

# The defaults defined in the Docker image are used.

2026

# - If you supply a command but no args:

2027

# The default EntryPoint and the default Cmd defined in the Docker image

2028

# are ignored. Your command is run without any arguments.

2029

# - If you supply only args:

2030

# The default Entrypoint defined in the Docker image is run with the args

2031

# that you supplied.

2032

# - If you supply a command and args:

2033

# The default Entrypoint and the default Cmd defined in the Docker image

2034

# are ignored. Your command is run with your args.

2035

# It cannot be set if custom container image is

2036

# not provided.

2037

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2038

# both cannot be set at the same time.

2039

"A String",

2040

],

2041

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

2042

# the one used in the custom container. This field is required if the replica

2043

# is a TPU worker that uses a custom container. Otherwise, do not specify

2044

# this field. This must be a [runtime version that currently supports

2045

# training with

2046

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

2047

#

2048

# Note that the version of TensorFlow included in a runtime version may

2049

# differ from the numbering of the runtime version itself, because it may

2050

# have a different [patch

2051

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

2052

# In this field, you must specify the runtime version (TensorFlow minor

2053

# version). For example, if your custom container runs TensorFlow `1.x.y`,

2054

# specify `1.x`.

2055

"containerCommand": [ # The command with which the replica's custom container is run.

2056

# If provided, it will override default ENTRYPOINT of the docker image.

2057

# If not provided, the docker image's ENTRYPOINT is used.

2058

# It cannot be set if custom container image is

2059

# not provided.

2060

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2061

# both cannot be set at the same time.

"A String",

],

},

"evaluatorCount": "A String", # Optional. The number of evaluator replicas to use for the training job.

2066

# Each replica in the cluster will be of the type specified in

2067

# `evaluator_type`.

2068

#

2069

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

2070

# set this value, you must also set `evaluator_type`.

2071

#

2072

# The default value is zero.

2073

"args": [ # Optional. Command-line arguments passed to the training application when it

2074

# starts. If your job uses a custom container, then the arguments are passed

2075

# to the container's <a class="external" target="_blank"

2076

# href="https://docs.docker.com/engine/reference/builder/#entrypoint">

2077

# `ENTRYPOINT`</a> command.

2078

"A String",

2079

],

2080

"pythonModule": "A String", # Required. The Python module name to run after installing the packages.

2081

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for training. You must

2082

# either specify this field or specify `masterConfig.imageUri`.

2083

#

2084

# For more information, see the [runtime version

2085

# list](/ai-platform/training/docs/runtime-version-list) and learn [how to

2086

# manage runtime versions](/ai-platform/training/docs/versioning).

2087

"parameterServerCount": "A String", # Optional. The number of parameter server replicas to use for the training

2088

# job. Each replica in the cluster will be of the type specified in

2089

# `parameter_server_type`.

2090

#

2091

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

2092

# set this value, you must also set `parameter_server_type`.

2093

#

2094

# The default value is zero.

2095

"evaluatorConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for evaluators.

2096

#

2097

# You should only set `evaluatorConfig.acceleratorConfig` if

2098

# `evaluatorType` is set to a Compute Engine machine type. [Learn

2099

# about restrictions on accelerator configurations for

2100

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

2101

#

2102

# Set `evaluatorConfig.imageUri` only if you build a custom image for

2103

# your evaluator. If `evaluatorConfig.imageUri` has not been

2104

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

2105

# containers](/ai-platform/training/docs/distributed-training-containers).

2106

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

2107

# [Learn about restrictions on accelerator configurations for

2108

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

2109

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

2110

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

2111

# [accelerators for online

2112

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

2113

"type": "A String", # The type of accelerator to use.

2114

"count": "A String", # The number of accelerators to attach to each machine running the job.

2115

},

2116

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

2117

# Registry. Learn more about [configuring custom

2118

# containers](/ai-platform/training/docs/distributed-training-containers).

2119

"containerArgs": [ # Arguments to the entrypoint command.

2120

# The following rules apply for container_command and container_args:

2121

# - If you do not supply command or args:

2122

# The defaults defined in the Docker image are used.

2123

# - If you supply a command but no args:

2124

# The default EntryPoint and the default Cmd defined in the Docker image

2125

# are ignored. Your command is run without any arguments.

2126

# - If you supply only args:

2127

# The default Entrypoint defined in the Docker image is run with the args

2128

# that you supplied.

2129

# - If you supply a command and args:

2130

# The default Entrypoint and the default Cmd defined in the Docker image

2131

# are ignored. Your command is run with your args.

2132

# It cannot be set if custom container image is

2133

# not provided.

2134

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2135

# both cannot be set at the same time.

2136

"A String",

2137

],

2138

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

2139

# the one used in the custom container. This field is required if the replica

2140

# is a TPU worker that uses a custom container. Otherwise, do not specify

2141

# this field. This must be a [runtime version that currently supports

2142

# training with

2143

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

2144

#

2145

# Note that the version of TensorFlow included in a runtime version may

2146

# differ from the numbering of the runtime version itself, because it may

2147

# have a different [patch

2148

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

2149

# In this field, you must specify the runtime version (TensorFlow minor

2150

# version). For example, if your custom container runs TensorFlow `1.x.y`,

2151

# specify `1.x`.

2152

"containerCommand": [ # The command with which the replica's custom container is run.

2153

# If provided, it will override default ENTRYPOINT of the docker image.

2154

# If not provided, the docker image's ENTRYPOINT is used.

2155

# It cannot be set if custom container image is

2156

# not provided.

2157

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2158

# both cannot be set at the same time.

"A String",

],

},

"encryptionConfig": { # Represents a custom encryption key configuration that can be applied to # Optional. Options for using customer-managed encryption keys (CMEK) to

2163

# protect resources created by a training job, instead of using Google's

2164

# default encryption. If this is set, then all resources created by the

2165

# training job will be encrypted with the customer-managed encryption key

2166

# that you specify.

2167

#

2168

# [Learn how and when to use CMEK with AI Platform

2169

# Training](/ai-platform/training/docs/cmek).

2170

# a resource.

2171

"kmsKeyName": "A String", # The Cloud KMS resource identifier of the customer-managed encryption key

2172

# used to protect a resource, such as a training job. It has the following

2173

# format:

2174

# `projects/{PROJECT_ID}/locations/{REGION}/keyRings/{KEY_RING_NAME}/cryptoKeys/{KEY_NAME}`

2175

},

2176

"workerCount": "A String", # Optional. The number of worker replicas to use for the training job. Each

2177

# replica in the cluster will be of the type specified in `worker_type`.

2178

#

2179

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

2180

# set this value, you must also set `worker_type`.

2181

#

2182

# The default value is zero.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2183

"scheduling": { # All parameters related to scheduling of training jobs. # Optional. Scheduling options for a training job.

2184

"maxWaitTime": "A String",

2185

"maxRunningTime": "A String", # Optional. The maximum job running time, expressed in seconds. The field can

2186

# contain up to nine fractional digits, terminated by `s`. If not specified,

2187

# this field defaults to `604800s` (seven days).

2188

#

2189

# If the training job is still running after this duration, AI Platform

2190

# Training cancels it.

2191

#

2192

# For example, if you want to ensure your job runs for no more than 2 hours,

2193

# set this field to `7200s` (2 hours * 60 minutes / hour * 60 seconds /

2194

# minute).

2195

#

2196

# If you submit your training job using the `gcloud` tool, you can [provide

2197

# this field in a `config.yaml`

2198

# file](/ai-platform/training/docs/training-jobs#formatting_your_configuration_parameters).

# For example:

#

# ```yaml

# trainingInput:

# ...

# scheduling:

# maxRunningTime: 7200s

2206

# ...

2207

# ```

2208

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2209

"scaleTier": "A String", # Required. Specifies the machine types, the number of replicas for workers

2210

# and parameter servers.

2211

"packageUris": [ # Required. The Google Cloud Storage location of the packages with

2212

# the training program and any additional dependencies.

2213

# The maximum number of package URIs is 100.

2214

"A String",

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2215

],

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2216

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2217

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

2218

# prevent simultaneous updates of a job from overwriting each other.

2219

# It is strongly suggested that systems make use of the `etag` in the

2220

# read-modify-write cycle to perform job updates in order to avoid race

2221

# conditions: An `etag` is returned in the response to `GetJob`, and

2222

# systems are expected to put that etag in the request to `UpdateJob` to

2223

# ensure that their change will be applied to the same version of the job.

2224

"jobId": "A String", # Required. The user-specified id of the job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2225

}</pre>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

</div>

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2229

<code class="details" id="getIamPolicy">getIamPolicy(resource, options_requestedPolicyVersion=None, x__xgafv=None)</code>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2230

<pre>Gets the access control policy for a resource.

2231

Returns an empty policy if the resource exists and does not have a policy

2232

set.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

2233

2234

Args:

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2235

resource: string, REQUIRED: The resource for which the policy is being requested.

2236

See the operation documentation for the appropriate value for this field. (required)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2237

options_requestedPolicyVersion: integer, Optional. The policy format version to be returned.

2238

2239

Valid values are 0, 1, and 3. Requests specifying an invalid value will be

2240

rejected.

2241

2242

Requests for policies with any conditional bindings must specify version 3.

2243

Policies without any conditional bindings may specify any valid value or

2244

leave the field unset.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2245

2246

To learn which resources support conditions in their IAM policies, see the

2247

[IAM

2248

documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

2249

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

2256

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2257

{ # An Identity and Access Management (IAM) policy, which specifies access

2258

# controls for Google Cloud resources.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2259

#

2260

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2261

# A `Policy` is a collection of `bindings`. A `binding` binds one or more

2262

# `members` to a single `role`. Members can be user accounts, service accounts,

2263

# Google groups, and domains (such as G Suite). A `role` is a named list of

2264

# permissions; each `role` can be an IAM predefined role or a user-created

2265

# custom role.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2266

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2267

# For some types of Google Cloud resources, a `binding` can also specify a

2268

# `condition`, which is a logical expression that allows access to a resource

2269

# only if the expression evaluates to `true`. A condition can add constraints

2270

# based on attributes of the request, the resource, or both. To learn which

2271

# resources support conditions in their IAM policies, see the

2272

# [IAM documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2273

#

2274

# **JSON example:**

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2275

#

2276

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2277

# "bindings": [

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2278

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2279

# "role": "roles/resourcemanager.organizationAdmin",

2280

# "members": [

2281

# "user:mike@example.com",

2282

# "group:admins@example.com",

2283

# "domain:google.com",

2284

# "serviceAccount:my-project-id@appspot.gserviceaccount.com"

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2285

# ]

2286

# },

2287

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2288

# "role": "roles/resourcemanager.organizationViewer",

2289

# "members": [

2290

# "user:eve@example.com"

2291

# ],

2292

# "condition": {

2293

# "title": "expirable access",

2294

# "description": "Does not grant access after Sep 2020",

2295

# "expression": "request.time < timestamp('2020-10-01T00:00:00.000Z')",

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2296

# }

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2297

# }

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2298

# ],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2299

# "etag": "BwWWja0YfJA=",

2300

# "version": 3

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2301

# }

2302

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2303

# **YAML example:**

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

#

# bindings:

# - members:

# - user:mike@example.com

2308

# - group:admins@example.com

2309

# - domain:google.com

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2310

# - serviceAccount:my-project-id@appspot.gserviceaccount.com

2311

# role: roles/resourcemanager.organizationAdmin

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2312

# - members:

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2313

# - user:eve@example.com

2314

# role: roles/resourcemanager.organizationViewer

2315

# condition:

2316

# title: expirable access

2317

# description: Does not grant access after Sep 2020

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2318

# expression: request.time < timestamp('2020-10-01T00:00:00.000Z')

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2319

# - etag: BwWWja0YfJA=

2320

# - version: 3

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2321

#

2322

# For a description of IAM and its features, see the

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2323

# [IAM documentation](https://cloud.google.com/iam/docs/).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2324

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

2325

# prevent simultaneous updates of a policy from overwriting each other.

2326

# It is strongly suggested that systems make use of the `etag` in the

2327

# read-modify-write cycle to perform policy updates in order to avoid race

2328

# conditions: An `etag` is returned in the response to `getIamPolicy`, and

2329

# systems are expected to put that etag in the request to `setIamPolicy` to

2330

# ensure that their change will be applied to the same version of the policy.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2331

#

2332

# **Important:** If you use IAM Conditions, you must include the `etag` field

2333

# whenever you call `setIamPolicy`. If you omit this field, then IAM allows

2334

# you to overwrite a version `3` policy with a version `1` policy, and all of

2335

# the conditions in the version `3` policy are lost.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2336

"auditConfigs": [ # Specifies cloud audit logging configuration for this policy.

2337

{ # Specifies the audit configuration for a service.

2338

# The configuration determines which permission types are logged, and what

2339

# identities, if any, are exempted from logging.

2340

# An AuditConfig must have one or more AuditLogConfigs.

2341

#

2342

# If there are AuditConfigs for both `allServices` and a specific service,

2343

# the union of the two AuditConfigs is used for that service: the log_types

2344

# specified in each AuditConfig are enabled, and the exempted_members in each

2345

# AuditLogConfig are exempted.

2346

#

2347

# Example Policy with multiple AuditConfigs:

#

# {

# "audit_configs": [

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2352

# "service": "allServices",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2353

# "audit_log_configs": [

2354

# {

2355

# "log_type": "DATA_READ",

2356

# "exempted_members": [

2357

# "user:jose@example.com"

2358

# ]

2359

# },

2360

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2361

# "log_type": "DATA_WRITE"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2362

# },

2363

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2364

# "log_type": "ADMIN_READ"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

# }

# ]

# },

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2369

# "service": "sampleservice.googleapis.com",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2370

# "audit_log_configs": [

2371

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2372

# "log_type": "DATA_READ"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2373

# },

2374

# {

2375

# "log_type": "DATA_WRITE",

2376

# "exempted_members": [

2377

# "user:aliya@example.com"

# ]

# }

# ]

# }

# ]

# }

#

# For sampleservice, this policy enables DATA_READ, DATA_WRITE and ADMIN_READ

2386

# logging. It also exempts jose@example.com from DATA_READ logging, and

2387

# aliya@example.com from DATA_WRITE logging.

2388

"service": "A String", # Specifies a service that will be enabled for audit logging.

2389

# For example, `storage.googleapis.com`, `cloudsql.googleapis.com`.

2390

# `allServices` is a special value that covers all services.

2391

"auditLogConfigs": [ # The configuration for logging of each type of permission.

2392

{ # Provides the configuration for logging a type of permissions.

# Example:

#

# {

# "audit_log_configs": [

2397

# {

2398

# "log_type": "DATA_READ",

2399

# "exempted_members": [

2400

# "user:jose@example.com"

2401

# ]

2402

# },

2403

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2404

# "log_type": "DATA_WRITE"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

# }

# ]

# }

#

# This enables 'DATA_READ' and 'DATA_WRITE' logging, while exempting

2410

# jose@example.com from DATA_READ logging.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2411

"logType": "A String", # The log type that this config enables.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2412

"exemptedMembers": [ # Specifies the identities that do not cause logging for this type of

2413

# permission.

2414

# Follows the same format of Binding.members.

2415

"A String",

2416

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

},

],

},

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2421

"version": 42, # Specifies the format of the policy.

2422

#

2423

# Valid values are `0`, `1`, and `3`. Requests that specify an invalid value

2424

# are rejected.

2425

#

2426

# Any operation that affects conditional role bindings must specify version

2427

# `3`. This requirement applies to the following operations:

2428

#

2429

# * Getting a policy that includes a conditional role binding

2430

# * Adding a conditional role binding to a policy

2431

# * Changing a conditional role binding in a policy

2432

# * Removing any role binding, with or without a condition, from a policy

2433

# that includes conditions

2434

#

2435

# **Important:** If you use IAM Conditions, you must include the `etag` field

2436

# whenever you call `setIamPolicy`. If you omit this field, then IAM allows

2437

# you to overwrite a version `3` policy with a version `1` policy, and all of

2438

# the conditions in the version `3` policy are lost.

2439

#

2440

# If a policy does not include any conditions, operations on that policy may

2441

# specify any valid version or leave the field unset.

2442

#

2443

# To learn which resources support conditions in their IAM policies, see the

2444

# [IAM documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2445

"bindings": [ # Associates a list of `members` to a `role`. Optionally, may specify a

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2446

# `condition` that determines how and when the `bindings` are applied. Each

2447

# of the `bindings` must contain at least one member.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2448

{ # Associates `members` with a `role`.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2449

"role": "A String", # Role that is assigned to `members`.

2450

# For example, `roles/viewer`, `roles/editor`, or `roles/owner`.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2451

"condition": { # Represents a textual expression in the Common Expression Language (CEL) # The condition that is associated with this binding.

2452

#

2453

# If the condition evaluates to `true`, then this binding applies to the

2454

# current request.

2455

#

2456

# If the condition evaluates to `false`, then this binding does not apply to

2457

# the current request. However, a different role binding might grant the same

2458

# role to one or more of the members in this binding.

2459

#

2460

# To learn which resources support conditions in their IAM policies, see the

2461

# [IAM

2462

# documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

2463

# syntax. CEL is a C-like expression language. The syntax and semantics of CEL

2464

# are documented at https://github.com/google/cel-spec.

2465

#

2466

# Example (Comparison):

2467

#

2468

# title: "Summary size limit"

2469

# description: "Determines if a summary is less than 100 chars"

2470

# expression: "document.summary.size() < 100"

2471

#

2472

# Example (Equality):

2473

#

2474

# title: "Requestor is owner"

2475

# description: "Determines if requestor is the document owner"

2476

# expression: "document.owner == request.auth.claims.email"

#

# Example (Logic):

#

# title: "Public documents"

2481

# description: "Determine whether the document should be publicly visible"

2482

# expression: "document.type != 'private' && document.type != 'internal'"

2483

#

2484

# Example (Data Manipulation):

2485

#

2486

# title: "Notification string"

2487

# description: "Create a notification string with a timestamp."

2488

# expression: "'New message received at ' + string(document.create_time)"

2489

#

2490

# The exact variables and functions that may be referenced within an expression

2491

# are determined by the service that evaluates it. See the service

2492

# documentation for additional information.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2493

"expression": "A String", # Textual representation of an expression in Common Expression Language

2494

# syntax.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2495

"title": "A String", # Optional. Title for the expression, i.e. a short string describing

2496

# its purpose. This can be used e.g. in UIs which allow to enter the

2497

# expression.

2498

"location": "A String", # Optional. String indicating the location of the expression for error

2499

# reporting, e.g. a file name and a position in the file.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2500

"description": "A String", # Optional. Description of the expression. This is a longer text which

2501

# describes the expression, e.g. when hovered over it in a UI.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2502

},

2503

"members": [ # Specifies the identities requesting access for a Cloud Platform resource.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2504

# `members` can have the following values:

2505

#

2506

# * `allUsers`: A special identifier that represents anyone who is

2507

# on the internet; with or without a Google account.

2508

#

2509

# * `allAuthenticatedUsers`: A special identifier that represents anyone

2510

# who is authenticated with a Google account or a service account.

2511

#

2512

# * `user:{emailid}`: An email address that represents a specific Google

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2513

# account. For example, `alice@example.com` .

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2514

#

2515

#

2516

# * `serviceAccount:{emailid}`: An email address that represents a service

2517

# account. For example, `my-other-app@appspot.gserviceaccount.com`.

2518

#

2519

# * `group:{emailid}`: An email address that represents a Google group.

2520

# For example, `admins@example.com`.

2521

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2522

# * `deleted:user:{emailid}?uid={uniqueid}`: An email address (plus unique

2523

# identifier) representing a user that has been recently deleted. For

2524

# example, `alice@example.com?uid=123456789012345678901`. If the user is

2525

# recovered, this value reverts to `user:{emailid}` and the recovered user

2526

# retains the role in the binding.

2527

#

2528

# * `deleted:serviceAccount:{emailid}?uid={uniqueid}`: An email address (plus

2529

# unique identifier) representing a service account that has been recently

2530

# deleted. For example,

2531

# `my-other-app@appspot.gserviceaccount.com?uid=123456789012345678901`.

2532

# If the service account is undeleted, this value reverts to

2533

# `serviceAccount:{emailid}` and the undeleted service account retains the

2534

# role in the binding.

2535

#

2536

# * `deleted:group:{emailid}?uid={uniqueid}`: An email address (plus unique

2537

# identifier) representing a Google group that has been recently

2538

# deleted. For example, `admins@example.com?uid=123456789012345678901`. If

2539

# the group is recovered, this value reverts to `group:{emailid}` and the

2540

# recovered group retains the role in the binding.

2541

#

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2542

#

2543

# * `domain:{domain}`: The G Suite domain (primary) that represents all the

2544

# users of that domain. For example, `google.com` or `example.com`.

2545

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2546

"A String",

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2547

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2548

},

2549

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2550

}</pre>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

</div>

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2554

<code class="details" id="list">list(parent, filter=None, pageToken=None, pageSize=None, x__xgafv=None)</code>

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

2555

<pre>Lists the jobs in the project.

2556

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2557

If there are no jobs that match the request parameters, the list

2558

request returns an empty response body: {}.

2559

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

2560

Args:

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2561

parent: string, Required. The name of the project for which to list jobs. (required)

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2562

filter: string, Optional. Specifies the subset of jobs to retrieve.

2563

You can filter on the value of one or more attributes of the job object.

2564

For example, retrieve jobs with a job identifier that starts with 'census':

2565

<p><code>gcloud ai-platform jobs list --filter='jobId:census*'</code>

2566

<p>List all failed jobs with names that start with 'rnn':

2567

<p><code>gcloud ai-platform jobs list --filter='jobId:rnn*

2568

AND state:FAILED'</code>

2569

<p>For more examples, see the guide to

2570

<a href="/ml-engine/docs/tensorflow/monitor-training">monitoring jobs</a>.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2571

pageToken: string, Optional. A page token to request the next page of results.

2572

2573

You get the token from the `next_page_token` field of the response from

2574

the previous call.

2575

pageSize: integer, Optional. The number of jobs to retrieve per "page" of results. If there

2576

are more remaining results than this number, the response message will

2577

contain a valid value in the `next_page_token` field.

2578

2579

The default value is 20, and the maximum page size is 100.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

2580

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

2587

2588

{ # Response message for the ListJobs method.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2589

"nextPageToken": "A String", # Optional. Pass this token as the `page_token` field of the request for a

2590

# subsequent call.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2591

"jobs": [ # The list of jobs.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

2592

{ # Represents a training or prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2593

"createTime": "A String", # Output only. When the job was created.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2594

"predictionInput": { # Represents input parameters for a prediction job. # Input parameters to create a prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2595

"dataFormat": "A String", # Required. The format of the input data files.

2596

"outputPath": "A String", # Required. The output Google Cloud Storage location.

2597

"inputPaths": [ # Required. The Cloud Storage location of the input data files. May contain

2598

# <a href="/storage/docs/gsutil/addlhelp/WildcardNames">wildcards</a>.

2599

"A String",

2600

],

2601

"region": "A String", # Required. The Google Compute Engine region to run the prediction job in.

2602

# See the <a href="/ml-engine/docs/tensorflow/regions">available regions</a>

2603

# for AI Platform services.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2604

"versionName": "A String", # Use this field if you want to specify a version of the model to use. The

2605

# string is formatted the same way as `model_version`, with the addition

2606

# of the version information:

2607

#

2608

# `"projects/YOUR_PROJECT/models/YOUR_MODEL/versions/YOUR_VERSION"`

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2609

"uri": "A String", # Use this field if you want to specify a Google Cloud Storage path for

2610

# the model to use.

2611

"outputDataFormat": "A String", # Optional. Format of the output data files, defaults to JSON.

2612

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for this batch

2613

# prediction. If not set, AI Platform will pick the runtime version used

2614

# during the CreateVersion request for this model version, or choose the

2615

# latest stable version when model version information is not available

2616

# such as when the model is specified by uri.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2617

"modelName": "A String", # Use this field if you want to use the default version for the specified

2618

# model. The string must use the following format:

2619

#

2620

# `"projects/YOUR_PROJECT/models/YOUR_MODEL"`

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2621

"signatureName": "A String", # Optional. The name of the signature defined in the SavedModel to use for

2622

# this job. Please refer to

2623

# [SavedModel](https://tensorflow.github.io/serving/serving_basic.html)

2624

# for information about how to use signatures.

2625

#

2626

# Defaults to

2627

# [DEFAULT_SERVING_SIGNATURE_DEF_KEY](https://www.tensorflow.org/api_docs/python/tf/saved_model/signature_constants)

2628

# , which is "serving_default".

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2629

"batchSize": "A String", # Optional. Number of records per batch, defaults to 64.

2630

# The service will buffer batch_size number of records in memory before

2631

# invoking one Tensorflow prediction call internally. So take the record

2632

# size and memory available into consideration when setting this parameter.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2633

"maxWorkerCount": "A String", # Optional. The maximum number of workers to be used for parallel processing.

2634

# Defaults to 10 if not specified.

2635

},

2636

"labels": { # Optional. One or more labels that you can add, to organize your jobs.

2637

# Each label is a key-value pair, where both the key and the value are

2638

# arbitrary strings that you supply.

2639

# For more information, see the documentation on

2640

# <a href="/ml-engine/docs/tensorflow/resource-labels">using labels</a>.

2641

"a_key": "A String",

2642

},

2643

"endTime": "A String", # Output only. When the job processing was completed.

2644

"trainingOutput": { # Represents results of a training job. Output only. # The current training job result.

2645

"trials": [ # Results for individual Hyperparameter trials.

2646

# Only set for hyperparameter tuning jobs.

2647

{ # Represents the result of a single hyperparameter tuning trial from a

2648

# training job. The TrainingOutput object that is returned on successful

2649

# completion of a training job with hyperparameter tuning includes a list

2650

# of HyperparameterOutput objects, one for each successful trial.

2651

"endTime": "A String", # Output only. End time for the trial.

2652

"finalMetric": { # An observed value of a metric. # The final objective metric seen for this trial.

2653

"trainingStep": "A String", # The global training step for this metric.

2654

"objectiveValue": 3.14, # The objective value at this training step.

2655

},

2656

"hyperparameters": { # The hyperparameters given to this trial.

2657

"a_key": "A String",

2658

},

2659

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

2660

# Only set for trials of built-in algorithms jobs that have succeeded.

2661

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

2662

# saves the trained model. Only set for successful jobs that don't use

2663

# hyperparameter tuning.

2664

"framework": "A String", # Framework on which the built-in algorithm was trained.

2665

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

2666

# trained.

2667

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

2668

},

2669

"startTime": "A String", # Output only. Start time for the trial.

2670

"allMetrics": [ # All recorded object metrics for this trial. This field is not currently

2671

# populated.

2672

{ # An observed value of a metric.

2673

"trainingStep": "A String", # The global training step for this metric.

2674

"objectiveValue": 3.14, # The objective value at this training step.

2675

},

2676

],

2677

"trialId": "A String", # The trial id for these results.

2678

"isTrialStoppedEarly": True or False, # True if the trial is stopped early.

2679

"state": "A String", # Output only. The detailed state of the trial.

2680

},

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2681

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2682

"completedTrialCount": "A String", # The number of hyperparameter tuning trials that completed successfully.

2683

# Only set for hyperparameter tuning jobs.

2684

"isHyperparameterTuningJob": True or False, # Whether this job is a hyperparameter tuning job.

2685

"isBuiltInAlgorithmJob": True or False, # Whether this job is a built-in Algorithm job.

2686

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

2687

# Only set for built-in algorithms jobs.

2688

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

2689

# saves the trained model. Only set for successful jobs that don't use

2690

# hyperparameter tuning.

2691

"framework": "A String", # Framework on which the built-in algorithm was trained.

2692

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

2693

# trained.

2694

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

2695

},

2696

"consumedMLUnits": 3.14, # The amount of ML units consumed by the job.

2697

"hyperparameterMetricTag": "A String", # The TensorFlow summary tag name used for optimizing hyperparameter tuning

2698

# trials. See

2699

# [`HyperparameterSpec.hyperparameterMetricTag`](#HyperparameterSpec.FIELDS.hyperparameter_metric_tag)

2700

# for more information. Only set for hyperparameter tuning jobs.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2701

},

2702

"errorMessage": "A String", # Output only. The details of a failure or a cancellation.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2703

"predictionOutput": { # Represents results of a prediction job. # The current prediction job result.

2704

"errorCount": "A String", # The number of data instances which resulted in errors.

2705

"nodeHours": 3.14, # Node hours used by the batch prediction job.

2706

"outputPath": "A String", # The output Google Cloud Storage location provided at the job creation time.

2707

"predictionCount": "A String", # The number of generated predictions.

2708

},

2709

"startTime": "A String", # Output only. When the job processing was started.

2710

"state": "A String", # Output only. The detailed state of a job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2711

"trainingInput": { # Represents input parameters for a training job. When using the gcloud command # Input parameters to create a training job.

2712

# to submit your training job, you can specify the input parameters as

2713

# command-line arguments and/or in a YAML configuration file referenced from

2714

# the --config command-line argument. For details, see the guide to [submitting

2715

# a training job](/ai-platform/training/docs/training-jobs).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2716

"serviceAccount": "A String", # Optional. Specifies the service account for workload run-as account.

2717

# Users submitting jobs must have act-as permission on this run-as account.

2718

# If not specified, then CMLE P4SA will be used by default.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2719

"workerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for workers.

2720

#

2721

# You should only set `workerConfig.acceleratorConfig` if `workerType` is set

2722

# to a Compute Engine machine type. [Learn about restrictions on accelerator

2723

# configurations for

2724

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

2725

#

2726

# Set `workerConfig.imageUri` only if you build a custom image for your

2727

# worker. If `workerConfig.imageUri` has not been set, AI Platform uses

2728

# the value of `masterConfig.imageUri`. Learn more about [configuring custom

2729

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2730

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

2731

# [Learn about restrictions on accelerator configurations for

2732

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

2733

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

2734

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

2735

# [accelerators for online

2736

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

2737

"type": "A String", # The type of accelerator to use.

2738

"count": "A String", # The number of accelerators to attach to each machine running the job.

2739

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2740

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

2741

# Registry. Learn more about [configuring custom

2742

# containers](/ai-platform/training/docs/distributed-training-containers).

2743

"containerArgs": [ # Arguments to the entrypoint command.

2744

# The following rules apply for container_command and container_args:

2745

# - If you do not supply command or args:

2746

# The defaults defined in the Docker image are used.

2747

# - If you supply a command but no args:

2748

# The default EntryPoint and the default Cmd defined in the Docker image

2749

# are ignored. Your command is run without any arguments.

2750

# - If you supply only args:

2751

# The default Entrypoint defined in the Docker image is run with the args

2752

# that you supplied.

2753

# - If you supply a command and args:

2754

# The default Entrypoint and the default Cmd defined in the Docker image

2755

# are ignored. Your command is run with your args.

2756

# It cannot be set if custom container image is

2757

# not provided.

2758

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2759

# both cannot be set at the same time.

2760

"A String",

2761

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2762

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

2763

# the one used in the custom container. This field is required if the replica

2764

# is a TPU worker that uses a custom container. Otherwise, do not specify

2765

# this field. This must be a [runtime version that currently supports

2766

# training with

2767

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

2768

#

2769

# Note that the version of TensorFlow included in a runtime version may

2770

# differ from the numbering of the runtime version itself, because it may

2771

# have a different [patch

2772

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

2773

# In this field, you must specify the runtime version (TensorFlow minor

2774

# version). For example, if your custom container runs TensorFlow `1.x.y`,

2775

# specify `1.x`.

2776

"containerCommand": [ # The command with which the replica's custom container is run.

2777

# If provided, it will override default ENTRYPOINT of the docker image.

2778

# If not provided, the docker image's ENTRYPOINT is used.

2779

# It cannot be set if custom container image is

2780

# not provided.

2781

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2782

# both cannot be set at the same time.

2783

"A String",

2784

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

2785

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2786

"useChiefInTfConfig": True or False, # Optional. Use `chief` instead of `master` in the `TF_CONFIG` environment

2787

# variable when training with a custom container. Defaults to `false`. [Learn

2788

# more about this

2789

# field.](/ai-platform/training/docs/distributed-training-details#chief-versus-master)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2790

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2791

# This field has no effect for training jobs that don't use a custom

2792

# container.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2793

"masterType": "A String", # Optional. Specifies the type of virtual machine to use for your training

2794

# job's master worker. You must specify this field when `scaleTier` is set to

2795

# `CUSTOM`.

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2796

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2797

# You can use certain Compute Engine machine types directly in this field.

2798

# The following types are supported:

#

# - `n1-standard-4`

# - `n1-standard-8`

# - `n1-standard-16`

# - `n1-standard-32`

# - `n1-standard-64`

# - `n1-standard-96`

# - `n1-highmem-2`

# - `n1-highmem-4`

# - `n1-highmem-8`

# - `n1-highmem-16`

# - `n1-highmem-32`

# - `n1-highmem-64`

# - `n1-highmem-96`

# - `n1-highcpu-16`

# - `n1-highcpu-32`

# - `n1-highcpu-64`

# - `n1-highcpu-96`

#

# Learn more about [using Compute Engine machine

2819

# types](/ml-engine/docs/machine-types#compute-engine-machine-types).

2820

#

2821

# Alternatively, you can use the following legacy machine types:

#

# - `standard`

# - `large_model`

# - `complex_model_s`

2826

# - `complex_model_m`

2827

# - `complex_model_l`

2828

# - `standard_gpu`

2829

# - `complex_model_m_gpu`

2830

# - `complex_model_l_gpu`

2831

# - `standard_p100`

2832

# - `complex_model_m_p100`

2833

# - `standard_v100`

2834

# - `large_model_v100`

2835

# - `complex_model_m_v100`

2836

# - `complex_model_l_v100`

2837

#

2838

# Learn more about [using legacy machine

2839

# types](/ml-engine/docs/machine-types#legacy-machine-types).

2840

#

2841

# Finally, if you want to use a TPU for training, specify `cloud_tpu` in this

2842

# field. Learn more about the [special configuration options for training

2843

# with

2844

# TPUs](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2845

"parameterServerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for parameter servers.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2846

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2847

# You should only set `parameterServerConfig.acceleratorConfig` if

2848

# `parameterServerType` is set to a Compute Engine machine type. [Learn

2849

# about restrictions on accelerator configurations for

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2850

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

2851

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2852

# Set `parameterServerConfig.imageUri` only if you build a custom image for

2853

# your parameter server. If `parameterServerConfig.imageUri` has not been

2854

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2855

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2856

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

2857

# [Learn about restrictions on accelerator configurations for

2858

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

2859

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

2860

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

2861

# [accelerators for online

2862

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

2863

"type": "A String", # The type of accelerator to use.

2864

"count": "A String", # The number of accelerators to attach to each machine running the job.

2865

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2866

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

2867

# Registry. Learn more about [configuring custom

2868

# containers](/ai-platform/training/docs/distributed-training-containers).

2869

"containerArgs": [ # Arguments to the entrypoint command.

2870

# The following rules apply for container_command and container_args:

2871

# - If you do not supply command or args:

2872

# The defaults defined in the Docker image are used.

2873

# - If you supply a command but no args:

2874

# The default EntryPoint and the default Cmd defined in the Docker image

2875

# are ignored. Your command is run without any arguments.

2876

# - If you supply only args:

2877

# The default Entrypoint defined in the Docker image is run with the args

2878

# that you supplied.

2879

# - If you supply a command and args:

2880

# The default Entrypoint and the default Cmd defined in the Docker image

2881

# are ignored. Your command is run with your args.

2882

# It cannot be set if custom container image is

2883

# not provided.

2884

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2885

# both cannot be set at the same time.

2886

"A String",

2887

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

2888

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

2889

# the one used in the custom container. This field is required if the replica

2890

# is a TPU worker that uses a custom container. Otherwise, do not specify

2891

# this field. This must be a [runtime version that currently supports

2892

# training with

2893

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

2894

#

2895

# Note that the version of TensorFlow included in a runtime version may

2896

# differ from the numbering of the runtime version itself, because it may

2897

# have a different [patch

2898

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

2899

# In this field, you must specify the runtime version (TensorFlow minor

2900

# version). For example, if your custom container runs TensorFlow `1.x.y`,

2901

# specify `1.x`.

2902

"containerCommand": [ # The command with which the replica's custom container is run.

2903

# If provided, it will override default ENTRYPOINT of the docker image.

2904

# If not provided, the docker image's ENTRYPOINT is used.

2905

# It cannot be set if custom container image is

2906

# not provided.

2907

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

2908

# both cannot be set at the same time.

2909

"A String",

2910

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2911

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2912

"region": "A String", # Required. The region to run the training job in. See the [available

2913

# regions](/ai-platform/training/docs/regions) for AI Platform Training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

2914

"jobDir": "A String", # Optional. A Google Cloud Storage path in which to store training outputs

2915

# and other data needed for training. This path is passed to your TensorFlow

2916

# program as the '--job-dir' command-line argument. The benefit of specifying

2917

# this field is that Cloud ML validates the path for use in training.

2918

"pythonVersion": "A String", # Optional. The version of Python used in training. You must either specify

2919

# this field or specify `masterConfig.imageUri`.

2920

#

2921

# The following Python versions are available:

2922

#

2923

# * Python '3.7' is available when `runtime_version` is set to '1.15' or

2924

# later.

2925

# * Python '3.5' is available when `runtime_version` is set to a version

2926

# from '1.4' to '1.14'.

2927

# * Python '2.7' is available when `runtime_version` is set to '1.15' or

2928

# earlier.

2929

#

2930

# Read more about the Python versions available for [each runtime

2931

# version](/ml-engine/docs/runtime-version-list).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

2932

"hyperparameters": { # Represents a set of hyperparameters to optimize. # Optional. The set of Hyperparameters to tune.

2933

"maxTrials": 42, # Optional. How many training trials should be attempted to optimize

2934

# the specified hyperparameters.

2935

#

2936

# Defaults to one.

2937

"enableTrialEarlyStopping": True or False, # Optional. Indicates if the hyperparameter tuning job enables auto trial

2938

# early stopping.

2939

"params": [ # Required. The set of parameters to tune.

2940

{ # Represents a single hyperparameter to optimize.

2941

"minValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

2942

# should be unset if type is `CATEGORICAL`. This value should be integers if

2943

# type is INTEGER.

2944

"categoricalValues": [ # Required if type is `CATEGORICAL`. The list of possible categories.

2945

"A String",

2946

],

2947

"scaleType": "A String", # Optional. How the parameter should be scaled to the hypercube.

2948

# Leave unset for categorical parameters.

2949

# Some kind of scaling is strongly recommended for real or integral

2950

# parameters (e.g., `UNIT_LINEAR_SCALE`).

2951

"discreteValues": [ # Required if type is `DISCRETE`.

2952

# A list of feasible points.

2953

# The list should be in strictly increasing order. For instance, this

2954

# parameter might have possible settings of 1.5, 2.5, and 4.0. This list

2955

# should not contain more than 1,000 values.

2956

3.14,

2957

],

2958

"type": "A String", # Required. The type of the parameter.

2959

"maxValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

2960

# should be unset if type is `CATEGORICAL`. This value should be integers if

2961

# type is `INTEGER`.

2962

"parameterName": "A String", # Required. The parameter name must be unique amongst all ParameterConfigs in

2963

# a HyperparameterSpec message. E.g., "learning_rate".

2964

},

2965

],

2966

"maxFailedTrials": 42, # Optional. The number of failed trials that need to be seen before failing

2967

# the hyperparameter tuning job. You can specify this field to override the

2968

# default failing criteria for AI Platform hyperparameter tuning jobs.

2969

#

2970

# Defaults to zero, which means the service decides when a hyperparameter

2971

# job should fail.

2972

"hyperparameterMetricTag": "A String", # Optional. The TensorFlow summary tag name to use for optimizing trials. For

2973

# current versions of TensorFlow, this tag name should exactly match what is

2974

# shown in TensorBoard, including all scopes. For versions of TensorFlow

2975

# prior to 0.12, this should be only the tag passed to tf.Summary.

2976

# By default, "training/hptuning/metric" will be used.

2977

"resumePreviousJobId": "A String", # Optional. The prior hyperparameter tuning job id that users hope to

2978

# continue with. The job id will be used to find the corresponding vizier

2979

# study guid and resume the study.

2980

"goal": "A String", # Required. The type of goal to use for tuning. Available types are

2981

# `MAXIMIZE` and `MINIMIZE`.

2982

#

2983

# Defaults to `MAXIMIZE`.

2984

"algorithm": "A String", # Optional. The search algorithm specified for the hyperparameter

2985

# tuning job.

2986

# Uses the default AI Platform hyperparameter tuning

2987

# algorithm if unspecified.

2988

"maxParallelTrials": 42, # Optional. The number of training trials to run concurrently.

2989

# You can reduce the time it takes to perform hyperparameter tuning by adding

2990

# trials in parallel. However, each trail only benefits from the information

2991

# gained in completed trials. That means that a trial does not get access to

2992

# the results of trials running at the same time, which could reduce the

2993

# quality of the overall optimization.

2994

#

2995

# Each trial will use the same scale tier and machine types.

#

# Defaults to one.

},

"evaluatorType": "A String", # Optional. Specifies the type of virtual machine to use for your training

3000

# job's evaluator nodes.

3001

#

3002

# The supported values are the same as those described in the entry for

3003

# `masterType`.

3004

#

3005

# This value must be consistent with the category of machine type that

3006

# `masterType` uses. In other words, both must be Compute Engine machine

3007

# types or both must be legacy machine types.

3008

#

3009

# This value must be present when `scaleTier` is set to `CUSTOM` and

3010

# `evaluatorCount` is greater than zero.

3011

"network": "A String", # Optional. The full name of the Google Compute Engine

3012

# [network](/compute/docs/networks-and-firewalls#networks) to which the Job

3013

# is peered. For example, projects/12345/global/networks/myVPC. Format is of

3014

# the form projects/{project}/global/networks/{network}. Where {project} is a

3015

# project number, as in '12345', and {network} is network name.".

3016

#

3017

# Private services access must already be configured for the network. If left

3018

# unspecified, the Job is not peered with any network. Learn more -

3019

# Connecting Job to user network over private

3020

# IP.

3021

"parameterServerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

3022

# job's parameter server.

3023

#

3024

# The supported values are the same as those described in the entry for

3025

# `master_type`.

3026

#

3027

# This value must be consistent with the category of machine type that

3028

# `masterType` uses. In other words, both must be Compute Engine machine

3029

# types or both must be legacy machine types.

3030

#

3031

# This value must be present when `scaleTier` is set to `CUSTOM` and

3032

# `parameter_server_count` is greater than zero.

3033

"workerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

3034

# job's worker nodes.

3035

#

3036

# The supported values are the same as those described in the entry for

3037

# `masterType`.

3038

#

3039

# This value must be consistent with the category of machine type that

3040

# `masterType` uses. In other words, both must be Compute Engine machine

3041

# types or both must be legacy machine types.

3042

#

3043

# If you use `cloud_tpu` for this value, see special instructions for

3044

# [configuring a custom TPU

3045

# machine](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

3046

#

3047

# This value must be present when `scaleTier` is set to `CUSTOM` and

3048

# `workerCount` is greater than zero.

3049

"masterConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for your master worker.

3050

#

3051

# You should only set `masterConfig.acceleratorConfig` if `masterType` is set

3052

# to a Compute Engine machine type. Learn about [restrictions on accelerator

3053

# configurations for

3054

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3055

#

3056

# Set `masterConfig.imageUri` only if you build a custom image. Only one of

3057

# `masterConfig.imageUri` and `runtimeVersion` should be set. Learn more

3058

# about [configuring custom

3059

# containers](/ai-platform/training/docs/distributed-training-containers).

3060

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

3061

# [Learn about restrictions on accelerator configurations for

3062

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3063

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

3064

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

3065

# [accelerators for online

3066

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

3067

"type": "A String", # The type of accelerator to use.

3068

"count": "A String", # The number of accelerators to attach to each machine running the job.

3069

},

3070

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

3071

# Registry. Learn more about [configuring custom

3072

# containers](/ai-platform/training/docs/distributed-training-containers).

3073

"containerArgs": [ # Arguments to the entrypoint command.

3074

# The following rules apply for container_command and container_args:

3075

# - If you do not supply command or args:

3076

# The defaults defined in the Docker image are used.

3077

# - If you supply a command but no args:

3078

# The default EntryPoint and the default Cmd defined in the Docker image

3079

# are ignored. Your command is run without any arguments.

3080

# - If you supply only args:

3081

# The default Entrypoint defined in the Docker image is run with the args

3082

# that you supplied.

3083

# - If you supply a command and args:

3084

# The default Entrypoint and the default Cmd defined in the Docker image

3085

# are ignored. Your command is run with your args.

3086

# It cannot be set if custom container image is

3087

# not provided.

3088

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3089

# both cannot be set at the same time.

3090

"A String",

3091

],

3092

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

3093

# the one used in the custom container. This field is required if the replica

3094

# is a TPU worker that uses a custom container. Otherwise, do not specify

3095

# this field. This must be a [runtime version that currently supports

3096

# training with

3097

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

3098

#

3099

# Note that the version of TensorFlow included in a runtime version may

3100

# differ from the numbering of the runtime version itself, because it may

3101

# have a different [patch

3102

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

3103

# In this field, you must specify the runtime version (TensorFlow minor

3104

# version). For example, if your custom container runs TensorFlow `1.x.y`,

3105

# specify `1.x`.

3106

"containerCommand": [ # The command with which the replica's custom container is run.

3107

# If provided, it will override default ENTRYPOINT of the docker image.

3108

# If not provided, the docker image's ENTRYPOINT is used.

3109

# It cannot be set if custom container image is

3110

# not provided.

3111

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3112

# both cannot be set at the same time.

"A String",

],

},

"evaluatorCount": "A String", # Optional. The number of evaluator replicas to use for the training job.

3117

# Each replica in the cluster will be of the type specified in

3118

# `evaluator_type`.

3119

#

3120

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

3121

# set this value, you must also set `evaluator_type`.

3122

#

3123

# The default value is zero.

3124

"args": [ # Optional. Command-line arguments passed to the training application when it

3125

# starts. If your job uses a custom container, then the arguments are passed

3126

# to the container's <a class="external" target="_blank"

3127

# href="https://docs.docker.com/engine/reference/builder/#entrypoint">

3128

# `ENTRYPOINT`</a> command.

3129

"A String",

3130

],

3131

"pythonModule": "A String", # Required. The Python module name to run after installing the packages.

3132

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for training. You must

3133

# either specify this field or specify `masterConfig.imageUri`.

3134

#

3135

# For more information, see the [runtime version

3136

# list](/ai-platform/training/docs/runtime-version-list) and learn [how to

3137

# manage runtime versions](/ai-platform/training/docs/versioning).

3138

"parameterServerCount": "A String", # Optional. The number of parameter server replicas to use for the training

3139

# job. Each replica in the cluster will be of the type specified in

3140

# `parameter_server_type`.

3141

#

3142

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

3143

# set this value, you must also set `parameter_server_type`.

3144

#

3145

# The default value is zero.

3146

"evaluatorConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for evaluators.

3147

#

3148

# You should only set `evaluatorConfig.acceleratorConfig` if

3149

# `evaluatorType` is set to a Compute Engine machine type. [Learn

3150

# about restrictions on accelerator configurations for

3151

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3152

#

3153

# Set `evaluatorConfig.imageUri` only if you build a custom image for

3154

# your evaluator. If `evaluatorConfig.imageUri` has not been

3155

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

3156

# containers](/ai-platform/training/docs/distributed-training-containers).

3157

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

3158

# [Learn about restrictions on accelerator configurations for

3159

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3160

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

3161

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

3162

# [accelerators for online

3163

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

3164

"type": "A String", # The type of accelerator to use.

3165

"count": "A String", # The number of accelerators to attach to each machine running the job.

3166

},

3167

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

3168

# Registry. Learn more about [configuring custom

3169

# containers](/ai-platform/training/docs/distributed-training-containers).

3170

"containerArgs": [ # Arguments to the entrypoint command.

3171

# The following rules apply for container_command and container_args:

3172

# - If you do not supply command or args:

3173

# The defaults defined in the Docker image are used.

3174

# - If you supply a command but no args:

3175

# The default EntryPoint and the default Cmd defined in the Docker image

3176

# are ignored. Your command is run without any arguments.

3177

# - If you supply only args:

3178

# The default Entrypoint defined in the Docker image is run with the args

3179

# that you supplied.

3180

# - If you supply a command and args:

3181

# The default Entrypoint and the default Cmd defined in the Docker image

3182

# are ignored. Your command is run with your args.

3183

# It cannot be set if custom container image is

3184

# not provided.

3185

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3186

# both cannot be set at the same time.

3187

"A String",

3188

],

3189

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

3190

# the one used in the custom container. This field is required if the replica

3191

# is a TPU worker that uses a custom container. Otherwise, do not specify

3192

# this field. This must be a [runtime version that currently supports

3193

# training with

3194

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

3195

#

3196

# Note that the version of TensorFlow included in a runtime version may

3197

# differ from the numbering of the runtime version itself, because it may

3198

# have a different [patch

3199

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

3200

# In this field, you must specify the runtime version (TensorFlow minor

3201

# version). For example, if your custom container runs TensorFlow `1.x.y`,

3202

# specify `1.x`.

3203

"containerCommand": [ # The command with which the replica's custom container is run.

3204

# If provided, it will override default ENTRYPOINT of the docker image.

3205

# If not provided, the docker image's ENTRYPOINT is used.

3206

# It cannot be set if custom container image is

3207

# not provided.

3208

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3209

# both cannot be set at the same time.

"A String",

],

},

"encryptionConfig": { # Represents a custom encryption key configuration that can be applied to # Optional. Options for using customer-managed encryption keys (CMEK) to

3214

# protect resources created by a training job, instead of using Google's

3215

# default encryption. If this is set, then all resources created by the

3216

# training job will be encrypted with the customer-managed encryption key

3217

# that you specify.

3218

#

3219

# [Learn how and when to use CMEK with AI Platform

3220

# Training](/ai-platform/training/docs/cmek).

3221

# a resource.

3222

"kmsKeyName": "A String", # The Cloud KMS resource identifier of the customer-managed encryption key

3223

# used to protect a resource, such as a training job. It has the following

3224

# format:

3225

# `projects/{PROJECT_ID}/locations/{REGION}/keyRings/{KEY_RING_NAME}/cryptoKeys/{KEY_NAME}`

3226

},

3227

"workerCount": "A String", # Optional. The number of worker replicas to use for the training job. Each

3228

# replica in the cluster will be of the type specified in `worker_type`.

3229

#

3230

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

3231

# set this value, you must also set `worker_type`.

3232

#

3233

# The default value is zero.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3234

"scheduling": { # All parameters related to scheduling of training jobs. # Optional. Scheduling options for a training job.

3235

"maxWaitTime": "A String",

3236

"maxRunningTime": "A String", # Optional. The maximum job running time, expressed in seconds. The field can

3237

# contain up to nine fractional digits, terminated by `s`. If not specified,

3238

# this field defaults to `604800s` (seven days).

3239

#

3240

# If the training job is still running after this duration, AI Platform

3241

# Training cancels it.

3242

#

3243

# For example, if you want to ensure your job runs for no more than 2 hours,

3244

# set this field to `7200s` (2 hours * 60 minutes / hour * 60 seconds /

3245

# minute).

3246

#

3247

# If you submit your training job using the `gcloud` tool, you can [provide

3248

# this field in a `config.yaml`

3249

# file](/ai-platform/training/docs/training-jobs#formatting_your_configuration_parameters).

# For example:

#

# ```yaml

# trainingInput:

# ...

# scheduling:

# maxRunningTime: 7200s

3257

# ...

3258

# ```

3259

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3260

"scaleTier": "A String", # Required. Specifies the machine types, the number of replicas for workers

3261

# and parameter servers.

3262

"packageUris": [ # Required. The Google Cloud Storage location of the packages with

3263

# the training program and any additional dependencies.

3264

# The maximum number of package URIs is 100.

3265

"A String",

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

3266

],

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3267

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3268

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

3269

# prevent simultaneous updates of a job from overwriting each other.

3270

# It is strongly suggested that systems make use of the `etag` in the

3271

# read-modify-write cycle to perform job updates in order to avoid race

3272

# conditions: An `etag` is returned in the response to `GetJob`, and

3273

# systems are expected to put that etag in the request to `UpdateJob` to

3274

# ensure that their change will be applied to the same version of the job.

3275

"jobId": "A String", # Required. The user-specified id of the job.

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

},

],

}</pre>

</div>

<code class="details" id="list_next">list_next(previous_request, previous_response)</code>

3283

<pre>Retrieves the next page of results.

3284

3285

Args:

3286

previous_request: The request for the previous page. (required)

3287

previous_response: The response from the request for the previous page. (required)

3288

3289

Returns:

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3290

A request object that you can call 'execute()' on to request the next

Sai Cheemalapati

2017-03-13 12:12:03 -0400

[diff] [blame]

3291

page. Returns None if there are no more items in the collection.

</pre>

</div>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

3295

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3296

<code class="details" id="patch">patch(name, body=None, updateMask=None, x__xgafv=None)</code>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

3297

<pre>Updates a specific job resource.

3298

3299

Currently the only supported fields to update are `labels`.

3300

3301

Args:

3302

name: string, Required. The job name. (required)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3303

body: object, The request body.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

3304

The object takes the form of:

3305

3306

{ # Represents a training or prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3307

"createTime": "A String", # Output only. When the job was created.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3308

"predictionInput": { # Represents input parameters for a prediction job. # Input parameters to create a prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3309

"dataFormat": "A String", # Required. The format of the input data files.

3310

"outputPath": "A String", # Required. The output Google Cloud Storage location.

3311

"inputPaths": [ # Required. The Cloud Storage location of the input data files. May contain

3312

# <a href="/storage/docs/gsutil/addlhelp/WildcardNames">wildcards</a>.

3313

"A String",

3314

],

3315

"region": "A String", # Required. The Google Compute Engine region to run the prediction job in.

3316

# See the <a href="/ml-engine/docs/tensorflow/regions">available regions</a>

3317

# for AI Platform services.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3318

"versionName": "A String", # Use this field if you want to specify a version of the model to use. The

3319

# string is formatted the same way as `model_version`, with the addition

3320

# of the version information:

3321

#

3322

# `"projects/YOUR_PROJECT/models/YOUR_MODEL/versions/YOUR_VERSION"`

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3323

"uri": "A String", # Use this field if you want to specify a Google Cloud Storage path for

3324

# the model to use.

3325

"outputDataFormat": "A String", # Optional. Format of the output data files, defaults to JSON.

3326

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for this batch

3327

# prediction. If not set, AI Platform will pick the runtime version used

3328

# during the CreateVersion request for this model version, or choose the

3329

# latest stable version when model version information is not available

3330

# such as when the model is specified by uri.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3331

"modelName": "A String", # Use this field if you want to use the default version for the specified

3332

# model. The string must use the following format:

3333

#

3334

# `"projects/YOUR_PROJECT/models/YOUR_MODEL"`

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3335

"signatureName": "A String", # Optional. The name of the signature defined in the SavedModel to use for

3336

# this job. Please refer to

3337

# [SavedModel](https://tensorflow.github.io/serving/serving_basic.html)

3338

# for information about how to use signatures.

3339

#

3340

# Defaults to

3341

# [DEFAULT_SERVING_SIGNATURE_DEF_KEY](https://www.tensorflow.org/api_docs/python/tf/saved_model/signature_constants)

3342

# , which is "serving_default".

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3343

"batchSize": "A String", # Optional. Number of records per batch, defaults to 64.

3344

# The service will buffer batch_size number of records in memory before

3345

# invoking one Tensorflow prediction call internally. So take the record

3346

# size and memory available into consideration when setting this parameter.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3347

"maxWorkerCount": "A String", # Optional. The maximum number of workers to be used for parallel processing.

3348

# Defaults to 10 if not specified.

3349

},

3350

"labels": { # Optional. One or more labels that you can add, to organize your jobs.

3351

# Each label is a key-value pair, where both the key and the value are

3352

# arbitrary strings that you supply.

3353

# For more information, see the documentation on

3354

# <a href="/ml-engine/docs/tensorflow/resource-labels">using labels</a>.

3355

"a_key": "A String",

3356

},

3357

"endTime": "A String", # Output only. When the job processing was completed.

3358

"trainingOutput": { # Represents results of a training job. Output only. # The current training job result.

3359

"trials": [ # Results for individual Hyperparameter trials.

3360

# Only set for hyperparameter tuning jobs.

3361

{ # Represents the result of a single hyperparameter tuning trial from a

3362

# training job. The TrainingOutput object that is returned on successful

3363

# completion of a training job with hyperparameter tuning includes a list

3364

# of HyperparameterOutput objects, one for each successful trial.

3365

"endTime": "A String", # Output only. End time for the trial.

3366

"finalMetric": { # An observed value of a metric. # The final objective metric seen for this trial.

3367

"trainingStep": "A String", # The global training step for this metric.

3368

"objectiveValue": 3.14, # The objective value at this training step.

3369

},

3370

"hyperparameters": { # The hyperparameters given to this trial.

3371

"a_key": "A String",

3372

},

3373

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

3374

# Only set for trials of built-in algorithms jobs that have succeeded.

3375

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

3376

# saves the trained model. Only set for successful jobs that don't use

3377

# hyperparameter tuning.

3378

"framework": "A String", # Framework on which the built-in algorithm was trained.

3379

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

3380

# trained.

3381

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

3382

},

3383

"startTime": "A String", # Output only. Start time for the trial.

3384

"allMetrics": [ # All recorded object metrics for this trial. This field is not currently

3385

# populated.

3386

{ # An observed value of a metric.

3387

"trainingStep": "A String", # The global training step for this metric.

3388

"objectiveValue": 3.14, # The objective value at this training step.

3389

},

3390

],

3391

"trialId": "A String", # The trial id for these results.

3392

"isTrialStoppedEarly": True or False, # True if the trial is stopped early.

3393

"state": "A String", # Output only. The detailed state of the trial.

3394

},

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3395

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3396

"completedTrialCount": "A String", # The number of hyperparameter tuning trials that completed successfully.

3397

# Only set for hyperparameter tuning jobs.

3398

"isHyperparameterTuningJob": True or False, # Whether this job is a hyperparameter tuning job.

3399

"isBuiltInAlgorithmJob": True or False, # Whether this job is a built-in Algorithm job.

3400

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

3401

# Only set for built-in algorithms jobs.

3402

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

3403

# saves the trained model. Only set for successful jobs that don't use

3404

# hyperparameter tuning.

3405

"framework": "A String", # Framework on which the built-in algorithm was trained.

3406

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

3407

# trained.

3408

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

3409

},

3410

"consumedMLUnits": 3.14, # The amount of ML units consumed by the job.

3411

"hyperparameterMetricTag": "A String", # The TensorFlow summary tag name used for optimizing hyperparameter tuning

3412

# trials. See

3413

# [`HyperparameterSpec.hyperparameterMetricTag`](#HyperparameterSpec.FIELDS.hyperparameter_metric_tag)

3414

# for more information. Only set for hyperparameter tuning jobs.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3415

},

3416

"errorMessage": "A String", # Output only. The details of a failure or a cancellation.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3417

"predictionOutput": { # Represents results of a prediction job. # The current prediction job result.

3418

"errorCount": "A String", # The number of data instances which resulted in errors.

3419

"nodeHours": 3.14, # Node hours used by the batch prediction job.

3420

"outputPath": "A String", # The output Google Cloud Storage location provided at the job creation time.

3421

"predictionCount": "A String", # The number of generated predictions.

3422

},

3423

"startTime": "A String", # Output only. When the job processing was started.

3424

"state": "A String", # Output only. The detailed state of a job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3425

"trainingInput": { # Represents input parameters for a training job. When using the gcloud command # Input parameters to create a training job.

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3426

# to submit your training job, you can specify the input parameters as

3427

# command-line arguments and/or in a YAML configuration file referenced from

3428

# the --config command-line argument. For details, see the guide to [submitting

3429

# a training job](/ai-platform/training/docs/training-jobs).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3430

"serviceAccount": "A String", # Optional. Specifies the service account for workload run-as account.

3431

# Users submitting jobs must have act-as permission on this run-as account.

3432

# If not specified, then CMLE P4SA will be used by default.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3433

"workerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for workers.

3434

#

3435

# You should only set `workerConfig.acceleratorConfig` if `workerType` is set

3436

# to a Compute Engine machine type. [Learn about restrictions on accelerator

3437

# configurations for

3438

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3439

#

3440

# Set `workerConfig.imageUri` only if you build a custom image for your

3441

# worker. If `workerConfig.imageUri` has not been set, AI Platform uses

3442

# the value of `masterConfig.imageUri`. Learn more about [configuring custom

3443

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3444

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

3445

# [Learn about restrictions on accelerator configurations for

3446

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3447

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

3448

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

3449

# [accelerators for online

3450

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

3451

"type": "A String", # The type of accelerator to use.

3452

"count": "A String", # The number of accelerators to attach to each machine running the job.

3453

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3454

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

3455

# Registry. Learn more about [configuring custom

3456

# containers](/ai-platform/training/docs/distributed-training-containers).

3457

"containerArgs": [ # Arguments to the entrypoint command.

3458

# The following rules apply for container_command and container_args:

3459

# - If you do not supply command or args:

3460

# The defaults defined in the Docker image are used.

3461

# - If you supply a command but no args:

3462

# The default EntryPoint and the default Cmd defined in the Docker image

3463

# are ignored. Your command is run without any arguments.

3464

# - If you supply only args:

3465

# The default Entrypoint defined in the Docker image is run with the args

3466

# that you supplied.

3467

# - If you supply a command and args:

3468

# The default Entrypoint and the default Cmd defined in the Docker image

3469

# are ignored. Your command is run with your args.

3470

# It cannot be set if custom container image is

3471

# not provided.

3472

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3473

# both cannot be set at the same time.

3474

"A String",

3475

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3476

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

3477

# the one used in the custom container. This field is required if the replica

3478

# is a TPU worker that uses a custom container. Otherwise, do not specify

3479

# this field. This must be a [runtime version that currently supports

3480

# training with

3481

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

3482

#

3483

# Note that the version of TensorFlow included in a runtime version may

3484

# differ from the numbering of the runtime version itself, because it may

3485

# have a different [patch

3486

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

3487

# In this field, you must specify the runtime version (TensorFlow minor

3488

# version). For example, if your custom container runs TensorFlow `1.x.y`,

3489

# specify `1.x`.

3490

"containerCommand": [ # The command with which the replica's custom container is run.

3491

# If provided, it will override default ENTRYPOINT of the docker image.

3492

# If not provided, the docker image's ENTRYPOINT is used.

3493

# It cannot be set if custom container image is

3494

# not provided.

3495

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3496

# both cannot be set at the same time.

3497

"A String",

3498

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3499

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3500

"useChiefInTfConfig": True or False, # Optional. Use `chief` instead of `master` in the `TF_CONFIG` environment

3501

# variable when training with a custom container. Defaults to `false`. [Learn

3502

# more about this

3503

# field.](/ai-platform/training/docs/distributed-training-details#chief-versus-master)

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3504

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3505

# This field has no effect for training jobs that don't use a custom

3506

# container.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3507

"masterType": "A String", # Optional. Specifies the type of virtual machine to use for your training

3508

# job's master worker. You must specify this field when `scaleTier` is set to

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3509

# `CUSTOM`.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

3510

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3511

# You can use certain Compute Engine machine types directly in this field.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

3512

# The following types are supported:

3513

#

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

# - `n1-standard-4`

# - `n1-standard-8`

# - `n1-standard-16`

# - `n1-standard-32`

# - `n1-standard-64`

# - `n1-standard-96`

# - `n1-highmem-2`

# - `n1-highmem-4`

# - `n1-highmem-8`

# - `n1-highmem-16`

# - `n1-highmem-32`

# - `n1-highmem-64`

# - `n1-highmem-96`

# - `n1-highcpu-16`

# - `n1-highcpu-32`

# - `n1-highcpu-64`

# - `n1-highcpu-96`

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3532

# Learn more about [using Compute Engine machine

3533

# types](/ml-engine/docs/machine-types#compute-engine-machine-types).

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

3534

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3535

# Alternatively, you can use the following legacy machine types:

#

# - `standard`

# - `large_model`

# - `complex_model_s`

3540

# - `complex_model_m`

3541

# - `complex_model_l`

3542

# - `standard_gpu`

3543

# - `complex_model_m_gpu`

3544

# - `complex_model_l_gpu`

3545

# - `standard_p100`

3546

# - `complex_model_m_p100`

3547

# - `standard_v100`

3548

# - `large_model_v100`

3549

# - `complex_model_m_v100`

3550

# - `complex_model_l_v100`

3551

#

3552

# Learn more about [using legacy machine

3553

# types](/ml-engine/docs/machine-types#legacy-machine-types).

3554

#

3555

# Finally, if you want to use a TPU for training, specify `cloud_tpu` in this

3556

# field. Learn more about the [special configuration options for training

3557

# with

3558

# TPUs](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3559

"parameterServerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for parameter servers.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3560

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3561

# You should only set `parameterServerConfig.acceleratorConfig` if

3562

# `parameterServerType` is set to a Compute Engine machine type. [Learn

3563

# about restrictions on accelerator configurations for

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3564

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3565

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3566

# Set `parameterServerConfig.imageUri` only if you build a custom image for

3567

# your parameter server. If `parameterServerConfig.imageUri` has not been

3568

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3569

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3570

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

3571

# [Learn about restrictions on accelerator configurations for

3572

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3573

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

3574

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

3575

# [accelerators for online

3576

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

3577

"type": "A String", # The type of accelerator to use.

3578

"count": "A String", # The number of accelerators to attach to each machine running the job.

3579

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3580

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

3581

# Registry. Learn more about [configuring custom

3582

# containers](/ai-platform/training/docs/distributed-training-containers).

3583

"containerArgs": [ # Arguments to the entrypoint command.

3584

# The following rules apply for container_command and container_args:

3585

# - If you do not supply command or args:

3586

# The defaults defined in the Docker image are used.

3587

# - If you supply a command but no args:

3588

# The default EntryPoint and the default Cmd defined in the Docker image

3589

# are ignored. Your command is run without any arguments.

3590

# - If you supply only args:

3591

# The default Entrypoint defined in the Docker image is run with the args

3592

# that you supplied.

3593

# - If you supply a command and args:

3594

# The default Entrypoint and the default Cmd defined in the Docker image

3595

# are ignored. Your command is run with your args.

3596

# It cannot be set if custom container image is

3597

# not provided.

3598

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3599

# both cannot be set at the same time.

3600

"A String",

3601

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

3602

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

3603

# the one used in the custom container. This field is required if the replica

3604

# is a TPU worker that uses a custom container. Otherwise, do not specify

3605

# this field. This must be a [runtime version that currently supports

3606

# training with

3607

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

3608

#

3609

# Note that the version of TensorFlow included in a runtime version may

3610

# differ from the numbering of the runtime version itself, because it may

3611

# have a different [patch

3612

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

3613

# In this field, you must specify the runtime version (TensorFlow minor

3614

# version). For example, if your custom container runs TensorFlow `1.x.y`,

3615

# specify `1.x`.

3616

"containerCommand": [ # The command with which the replica's custom container is run.

3617

# If provided, it will override default ENTRYPOINT of the docker image.

3618

# If not provided, the docker image's ENTRYPOINT is used.

3619

# It cannot be set if custom container image is

3620

# not provided.

3621

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3622

# both cannot be set at the same time.

3623

"A String",

3624

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3625

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3626

"region": "A String", # Required. The region to run the training job in. See the [available

3627

# regions](/ai-platform/training/docs/regions) for AI Platform Training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3628

"jobDir": "A String", # Optional. A Google Cloud Storage path in which to store training outputs

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3629

# and other data needed for training. This path is passed to your TensorFlow

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3630

# program as the '--job-dir' command-line argument. The benefit of specifying

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3631

# this field is that Cloud ML validates the path for use in training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3632

"pythonVersion": "A String", # Optional. The version of Python used in training. You must either specify

3633

# this field or specify `masterConfig.imageUri`.

3634

#

3635

# The following Python versions are available:

3636

#

3637

# * Python '3.7' is available when `runtime_version` is set to '1.15' or

3638

# later.

3639

# * Python '3.5' is available when `runtime_version` is set to a version

3640

# from '1.4' to '1.14'.

3641

# * Python '2.7' is available when `runtime_version` is set to '1.15' or

3642

# earlier.

3643

#

3644

# Read more about the Python versions available for [each runtime

3645

# version](/ml-engine/docs/runtime-version-list).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3646

"hyperparameters": { # Represents a set of hyperparameters to optimize. # Optional. The set of Hyperparameters to tune.

3647

"maxTrials": 42, # Optional. How many training trials should be attempted to optimize

3648

# the specified hyperparameters.

3649

#

3650

# Defaults to one.

3651

"enableTrialEarlyStopping": True or False, # Optional. Indicates if the hyperparameter tuning job enables auto trial

3652

# early stopping.

3653

"params": [ # Required. The set of parameters to tune.

3654

{ # Represents a single hyperparameter to optimize.

3655

"minValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

3656

# should be unset if type is `CATEGORICAL`. This value should be integers if

3657

# type is INTEGER.

3658

"categoricalValues": [ # Required if type is `CATEGORICAL`. The list of possible categories.

3659

"A String",

3660

],

3661

"scaleType": "A String", # Optional. How the parameter should be scaled to the hypercube.

3662

# Leave unset for categorical parameters.

3663

# Some kind of scaling is strongly recommended for real or integral

3664

# parameters (e.g., `UNIT_LINEAR_SCALE`).

3665

"discreteValues": [ # Required if type is `DISCRETE`.

3666

# A list of feasible points.

3667

# The list should be in strictly increasing order. For instance, this

3668

# parameter might have possible settings of 1.5, 2.5, and 4.0. This list

3669

# should not contain more than 1,000 values.

3670

3.14,

3671

],

3672

"type": "A String", # Required. The type of the parameter.

3673

"maxValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

3674

# should be unset if type is `CATEGORICAL`. This value should be integers if

3675

# type is `INTEGER`.

3676

"parameterName": "A String", # Required. The parameter name must be unique amongst all ParameterConfigs in

3677

# a HyperparameterSpec message. E.g., "learning_rate".

3678

},

3679

],

3680

"maxFailedTrials": 42, # Optional. The number of failed trials that need to be seen before failing

3681

# the hyperparameter tuning job. You can specify this field to override the

3682

# default failing criteria for AI Platform hyperparameter tuning jobs.

3683

#

3684

# Defaults to zero, which means the service decides when a hyperparameter

3685

# job should fail.

3686

"hyperparameterMetricTag": "A String", # Optional. The TensorFlow summary tag name to use for optimizing trials. For

3687

# current versions of TensorFlow, this tag name should exactly match what is

3688

# shown in TensorBoard, including all scopes. For versions of TensorFlow

3689

# prior to 0.12, this should be only the tag passed to tf.Summary.

3690

# By default, "training/hptuning/metric" will be used.

3691

"resumePreviousJobId": "A String", # Optional. The prior hyperparameter tuning job id that users hope to

3692

# continue with. The job id will be used to find the corresponding vizier

3693

# study guid and resume the study.

3694

"goal": "A String", # Required. The type of goal to use for tuning. Available types are

3695

# `MAXIMIZE` and `MINIMIZE`.

3696

#

3697

# Defaults to `MAXIMIZE`.

3698

"algorithm": "A String", # Optional. The search algorithm specified for the hyperparameter

3699

# tuning job.

3700

# Uses the default AI Platform hyperparameter tuning

3701

# algorithm if unspecified.

3702

"maxParallelTrials": 42, # Optional. The number of training trials to run concurrently.

3703

# You can reduce the time it takes to perform hyperparameter tuning by adding

3704

# trials in parallel. However, each trail only benefits from the information

3705

# gained in completed trials. That means that a trial does not get access to

3706

# the results of trials running at the same time, which could reduce the

3707

# quality of the overall optimization.

3708

#

3709

# Each trial will use the same scale tier and machine types.

#

# Defaults to one.

},

"evaluatorType": "A String", # Optional. Specifies the type of virtual machine to use for your training

3714

# job's evaluator nodes.

3715

#

3716

# The supported values are the same as those described in the entry for

3717

# `masterType`.

3718

#

3719

# This value must be consistent with the category of machine type that

3720

# `masterType` uses. In other words, both must be Compute Engine machine

3721

# types or both must be legacy machine types.

3722

#

3723

# This value must be present when `scaleTier` is set to `CUSTOM` and

3724

# `evaluatorCount` is greater than zero.

3725

"network": "A String", # Optional. The full name of the Google Compute Engine

3726

# [network](/compute/docs/networks-and-firewalls#networks) to which the Job

3727

# is peered. For example, projects/12345/global/networks/myVPC. Format is of

3728

# the form projects/{project}/global/networks/{network}. Where {project} is a

3729

# project number, as in '12345', and {network} is network name.".

3730

#

3731

# Private services access must already be configured for the network. If left

3732

# unspecified, the Job is not peered with any network. Learn more -

3733

# Connecting Job to user network over private

3734

# IP.

3735

"parameterServerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

3736

# job's parameter server.

3737

#

3738

# The supported values are the same as those described in the entry for

3739

# `master_type`.

3740

#

3741

# This value must be consistent with the category of machine type that

3742

# `masterType` uses. In other words, both must be Compute Engine machine

3743

# types or both must be legacy machine types.

3744

#

3745

# This value must be present when `scaleTier` is set to `CUSTOM` and

3746

# `parameter_server_count` is greater than zero.

3747

"workerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

3748

# job's worker nodes.

3749

#

3750

# The supported values are the same as those described in the entry for

3751

# `masterType`.

3752

#

3753

# This value must be consistent with the category of machine type that

3754

# `masterType` uses. In other words, both must be Compute Engine machine

3755

# types or both must be legacy machine types.

3756

#

3757

# If you use `cloud_tpu` for this value, see special instructions for

3758

# [configuring a custom TPU

3759

# machine](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

3760

#

3761

# This value must be present when `scaleTier` is set to `CUSTOM` and

3762

# `workerCount` is greater than zero.

3763

"masterConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for your master worker.

3764

#

3765

# You should only set `masterConfig.acceleratorConfig` if `masterType` is set

3766

# to a Compute Engine machine type. Learn about [restrictions on accelerator

3767

# configurations for

3768

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3769

#

3770

# Set `masterConfig.imageUri` only if you build a custom image. Only one of

3771

# `masterConfig.imageUri` and `runtimeVersion` should be set. Learn more

3772

# about [configuring custom

3773

# containers](/ai-platform/training/docs/distributed-training-containers).

3774

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

3775

# [Learn about restrictions on accelerator configurations for

3776

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3777

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

3778

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

3779

# [accelerators for online

3780

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

3781

"type": "A String", # The type of accelerator to use.

3782

"count": "A String", # The number of accelerators to attach to each machine running the job.

3783

},

3784

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

3785

# Registry. Learn more about [configuring custom

3786

# containers](/ai-platform/training/docs/distributed-training-containers).

3787

"containerArgs": [ # Arguments to the entrypoint command.

3788

# The following rules apply for container_command and container_args:

3789

# - If you do not supply command or args:

3790

# The defaults defined in the Docker image are used.

3791

# - If you supply a command but no args:

3792

# The default EntryPoint and the default Cmd defined in the Docker image

3793

# are ignored. Your command is run without any arguments.

3794

# - If you supply only args:

3795

# The default Entrypoint defined in the Docker image is run with the args

3796

# that you supplied.

3797

# - If you supply a command and args:

3798

# The default Entrypoint and the default Cmd defined in the Docker image

3799

# are ignored. Your command is run with your args.

3800

# It cannot be set if custom container image is

3801

# not provided.

3802

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3803

# both cannot be set at the same time.

3804

"A String",

3805

],

3806

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

3807

# the one used in the custom container. This field is required if the replica

3808

# is a TPU worker that uses a custom container. Otherwise, do not specify

3809

# this field. This must be a [runtime version that currently supports

3810

# training with

3811

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

3812

#

3813

# Note that the version of TensorFlow included in a runtime version may

3814

# differ from the numbering of the runtime version itself, because it may

3815

# have a different [patch

3816

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

3817

# In this field, you must specify the runtime version (TensorFlow minor

3818

# version). For example, if your custom container runs TensorFlow `1.x.y`,

3819

# specify `1.x`.

3820

"containerCommand": [ # The command with which the replica's custom container is run.

3821

# If provided, it will override default ENTRYPOINT of the docker image.

3822

# If not provided, the docker image's ENTRYPOINT is used.

3823

# It cannot be set if custom container image is

3824

# not provided.

3825

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3826

# both cannot be set at the same time.

"A String",

],

},

"evaluatorCount": "A String", # Optional. The number of evaluator replicas to use for the training job.

3831

# Each replica in the cluster will be of the type specified in

3832

# `evaluator_type`.

3833

#

3834

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

3835

# set this value, you must also set `evaluator_type`.

3836

#

3837

# The default value is zero.

3838

"args": [ # Optional. Command-line arguments passed to the training application when it

3839

# starts. If your job uses a custom container, then the arguments are passed

3840

# to the container's <a class="external" target="_blank"

3841

# href="https://docs.docker.com/engine/reference/builder/#entrypoint">

3842

# `ENTRYPOINT`</a> command.

3843

"A String",

3844

],

3845

"pythonModule": "A String", # Required. The Python module name to run after installing the packages.

3846

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for training. You must

3847

# either specify this field or specify `masterConfig.imageUri`.

3848

#

3849

# For more information, see the [runtime version

3850

# list](/ai-platform/training/docs/runtime-version-list) and learn [how to

3851

# manage runtime versions](/ai-platform/training/docs/versioning).

3852

"parameterServerCount": "A String", # Optional. The number of parameter server replicas to use for the training

3853

# job. Each replica in the cluster will be of the type specified in

3854

# `parameter_server_type`.

3855

#

3856

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

3857

# set this value, you must also set `parameter_server_type`.

3858

#

3859

# The default value is zero.

3860

"evaluatorConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for evaluators.

3861

#

3862

# You should only set `evaluatorConfig.acceleratorConfig` if

3863

# `evaluatorType` is set to a Compute Engine machine type. [Learn

3864

# about restrictions on accelerator configurations for

3865

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3866

#

3867

# Set `evaluatorConfig.imageUri` only if you build a custom image for

3868

# your evaluator. If `evaluatorConfig.imageUri` has not been

3869

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

3870

# containers](/ai-platform/training/docs/distributed-training-containers).

3871

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

3872

# [Learn about restrictions on accelerator configurations for

3873

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

3874

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

3875

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

3876

# [accelerators for online

3877

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

3878

"type": "A String", # The type of accelerator to use.

3879

"count": "A String", # The number of accelerators to attach to each machine running the job.

3880

},

3881

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

3882

# Registry. Learn more about [configuring custom

3883

# containers](/ai-platform/training/docs/distributed-training-containers).

3884

"containerArgs": [ # Arguments to the entrypoint command.

3885

# The following rules apply for container_command and container_args:

3886

# - If you do not supply command or args:

3887

# The defaults defined in the Docker image are used.

3888

# - If you supply a command but no args:

3889

# The default EntryPoint and the default Cmd defined in the Docker image

3890

# are ignored. Your command is run without any arguments.

3891

# - If you supply only args:

3892

# The default Entrypoint defined in the Docker image is run with the args

3893

# that you supplied.

3894

# - If you supply a command and args:

3895

# The default Entrypoint and the default Cmd defined in the Docker image

3896

# are ignored. Your command is run with your args.

3897

# It cannot be set if custom container image is

3898

# not provided.

3899

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3900

# both cannot be set at the same time.

3901

"A String",

3902

],

3903

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

3904

# the one used in the custom container. This field is required if the replica

3905

# is a TPU worker that uses a custom container. Otherwise, do not specify

3906

# this field. This must be a [runtime version that currently supports

3907

# training with

3908

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

3909

#

3910

# Note that the version of TensorFlow included in a runtime version may

3911

# differ from the numbering of the runtime version itself, because it may

3912

# have a different [patch

3913

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

3914

# In this field, you must specify the runtime version (TensorFlow minor

3915

# version). For example, if your custom container runs TensorFlow `1.x.y`,

3916

# specify `1.x`.

3917

"containerCommand": [ # The command with which the replica's custom container is run.

3918

# If provided, it will override default ENTRYPOINT of the docker image.

3919

# If not provided, the docker image's ENTRYPOINT is used.

3920

# It cannot be set if custom container image is

3921

# not provided.

3922

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

3923

# both cannot be set at the same time.

"A String",

],

},

"encryptionConfig": { # Represents a custom encryption key configuration that can be applied to # Optional. Options for using customer-managed encryption keys (CMEK) to

3928

# protect resources created by a training job, instead of using Google's

3929

# default encryption. If this is set, then all resources created by the

3930

# training job will be encrypted with the customer-managed encryption key

3931

# that you specify.

3932

#

3933

# [Learn how and when to use CMEK with AI Platform

3934

# Training](/ai-platform/training/docs/cmek).

3935

# a resource.

3936

"kmsKeyName": "A String", # The Cloud KMS resource identifier of the customer-managed encryption key

3937

# used to protect a resource, such as a training job. It has the following

3938

# format:

3939

# `projects/{PROJECT_ID}/locations/{REGION}/keyRings/{KEY_RING_NAME}/cryptoKeys/{KEY_NAME}`

3940

},

3941

"workerCount": "A String", # Optional. The number of worker replicas to use for the training job. Each

3942

# replica in the cluster will be of the type specified in `worker_type`.

3943

#

3944

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

3945

# set this value, you must also set `worker_type`.

3946

#

3947

# The default value is zero.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3948

"scheduling": { # All parameters related to scheduling of training jobs. # Optional. Scheduling options for a training job.

3949

"maxWaitTime": "A String",

3950

"maxRunningTime": "A String", # Optional. The maximum job running time, expressed in seconds. The field can

3951

# contain up to nine fractional digits, terminated by `s`. If not specified,

3952

# this field defaults to `604800s` (seven days).

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

3953

#

3954

# If the training job is still running after this duration, AI Platform

3955

# Training cancels it.

3956

#

3957

# For example, if you want to ensure your job runs for no more than 2 hours,

3958

# set this field to `7200s` (2 hours * 60 minutes / hour * 60 seconds /

3959

# minute).

3960

#

3961

# If you submit your training job using the `gcloud` tool, you can [provide

3962

# this field in a `config.yaml`

3963

# file](/ai-platform/training/docs/training-jobs#formatting_your_configuration_parameters).

# For example:

#

# ```yaml

# trainingInput:

# ...

# scheduling:

# maxRunningTime: 7200s

3971

# ...

3972

# ```

3973

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3974

"scaleTier": "A String", # Required. Specifies the machine types, the number of replicas for workers

3975

# and parameter servers.

3976

"packageUris": [ # Required. The Google Cloud Storage location of the packages with

3977

# the training program and any additional dependencies.

3978

# The maximum number of package URIs is 100.

3979

"A String",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3980

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

3981

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

3982

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

3983

# prevent simultaneous updates of a job from overwriting each other.

3984

# It is strongly suggested that systems make use of the `etag` in the

3985

# read-modify-write cycle to perform job updates in order to avoid race

3986

# conditions: An `etag` is returned in the response to `GetJob`, and

3987

# systems are expected to put that etag in the request to `UpdateJob` to

3988

# ensure that their change will be applied to the same version of the job.

3989

"jobId": "A String", # Required. The user-specified id of the job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

3990

}

3991

3992

updateMask: string, Required. Specifies the path, relative to `Job`, of the field to update.

3993

To adopt etag mechanism, include `etag` field in the mask, and include the

3994

`etag` value in your job resource.

3995

3996

For example, to change the labels of a job, the `update_mask` parameter

3997

would be specified as `labels`, `etag`, and the

3998

`PATCH` request body would specify the new value, as follows:

{

"labels": {

"owner": "Google",

"color": "Blue"

}

"etag": "33a64df551425fcc55e4d42a148795d9f25f89d4"

4005

}

4006

If `etag` matches the one on the server, the labels of the job will be

4007

replaced with the given ones, and the server end `etag` will be

4008

recalculated.

4009

4010

Currently the only supported update masks are `labels` and `etag`.

4011

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

4018

4019

{ # Represents a training or prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4020

"createTime": "A String", # Output only. When the job was created.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4021

"predictionInput": { # Represents input parameters for a prediction job. # Input parameters to create a prediction job.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4022

"dataFormat": "A String", # Required. The format of the input data files.

4023

"outputPath": "A String", # Required. The output Google Cloud Storage location.

4024

"inputPaths": [ # Required. The Cloud Storage location of the input data files. May contain

4025

# <a href="/storage/docs/gsutil/addlhelp/WildcardNames">wildcards</a>.

4026

"A String",

4027

],

4028

"region": "A String", # Required. The Google Compute Engine region to run the prediction job in.

4029

# See the <a href="/ml-engine/docs/tensorflow/regions">available regions</a>

4030

# for AI Platform services.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4031

"versionName": "A String", # Use this field if you want to specify a version of the model to use. The

4032

# string is formatted the same way as `model_version`, with the addition

4033

# of the version information:

4034

#

4035

# `"projects/YOUR_PROJECT/models/YOUR_MODEL/versions/YOUR_VERSION"`

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4036

"uri": "A String", # Use this field if you want to specify a Google Cloud Storage path for

4037

# the model to use.

4038

"outputDataFormat": "A String", # Optional. Format of the output data files, defaults to JSON.

4039

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for this batch

4040

# prediction. If not set, AI Platform will pick the runtime version used

4041

# during the CreateVersion request for this model version, or choose the

4042

# latest stable version when model version information is not available

4043

# such as when the model is specified by uri.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4044

"modelName": "A String", # Use this field if you want to use the default version for the specified

4045

# model. The string must use the following format:

4046

#

4047

# `"projects/YOUR_PROJECT/models/YOUR_MODEL"`

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4048

"signatureName": "A String", # Optional. The name of the signature defined in the SavedModel to use for

4049

# this job. Please refer to

4050

# [SavedModel](https://tensorflow.github.io/serving/serving_basic.html)

4051

# for information about how to use signatures.

4052

#

4053

# Defaults to

4054

# [DEFAULT_SERVING_SIGNATURE_DEF_KEY](https://www.tensorflow.org/api_docs/python/tf/saved_model/signature_constants)

4055

# , which is "serving_default".

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4056

"batchSize": "A String", # Optional. Number of records per batch, defaults to 64.

4057

# The service will buffer batch_size number of records in memory before

4058

# invoking one Tensorflow prediction call internally. So take the record

4059

# size and memory available into consideration when setting this parameter.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4060

"maxWorkerCount": "A String", # Optional. The maximum number of workers to be used for parallel processing.

4061

# Defaults to 10 if not specified.

4062

},

4063

"labels": { # Optional. One or more labels that you can add, to organize your jobs.

4064

# Each label is a key-value pair, where both the key and the value are

4065

# arbitrary strings that you supply.

4066

# For more information, see the documentation on

4067

# <a href="/ml-engine/docs/tensorflow/resource-labels">using labels</a>.

4068

"a_key": "A String",

4069

},

4070

"endTime": "A String", # Output only. When the job processing was completed.

4071

"trainingOutput": { # Represents results of a training job. Output only. # The current training job result.

4072

"trials": [ # Results for individual Hyperparameter trials.

4073

# Only set for hyperparameter tuning jobs.

4074

{ # Represents the result of a single hyperparameter tuning trial from a

4075

# training job. The TrainingOutput object that is returned on successful

4076

# completion of a training job with hyperparameter tuning includes a list

4077

# of HyperparameterOutput objects, one for each successful trial.

4078

"endTime": "A String", # Output only. End time for the trial.

4079

"finalMetric": { # An observed value of a metric. # The final objective metric seen for this trial.

4080

"trainingStep": "A String", # The global training step for this metric.

4081

"objectiveValue": 3.14, # The objective value at this training step.

4082

},

4083

"hyperparameters": { # The hyperparameters given to this trial.

4084

"a_key": "A String",

4085

},

4086

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

4087

# Only set for trials of built-in algorithms jobs that have succeeded.

4088

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

4089

# saves the trained model. Only set for successful jobs that don't use

4090

# hyperparameter tuning.

4091

"framework": "A String", # Framework on which the built-in algorithm was trained.

4092

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

4093

# trained.

4094

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

4095

},

4096

"startTime": "A String", # Output only. Start time for the trial.

4097

"allMetrics": [ # All recorded object metrics for this trial. This field is not currently

4098

# populated.

4099

{ # An observed value of a metric.

4100

"trainingStep": "A String", # The global training step for this metric.

4101

"objectiveValue": 3.14, # The objective value at this training step.

4102

},

4103

],

4104

"trialId": "A String", # The trial id for these results.

4105

"isTrialStoppedEarly": True or False, # True if the trial is stopped early.

4106

"state": "A String", # Output only. The detailed state of the trial.

4107

},

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4108

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4109

"completedTrialCount": "A String", # The number of hyperparameter tuning trials that completed successfully.

4110

# Only set for hyperparameter tuning jobs.

4111

"isHyperparameterTuningJob": True or False, # Whether this job is a hyperparameter tuning job.

4112

"isBuiltInAlgorithmJob": True or False, # Whether this job is a built-in Algorithm job.

4113

"builtInAlgorithmOutput": { # Represents output related to a built-in algorithm Job. # Details related to built-in algorithms jobs.

4114

# Only set for built-in algorithms jobs.

4115

"modelPath": "A String", # The Cloud Storage path to the `model/` directory where the training job

4116

# saves the trained model. Only set for successful jobs that don't use

4117

# hyperparameter tuning.

4118

"framework": "A String", # Framework on which the built-in algorithm was trained.

4119

"runtimeVersion": "A String", # AI Platform runtime version on which the built-in algorithm was

4120

# trained.

4121

"pythonVersion": "A String", # Python version on which the built-in algorithm was trained.

4122

},

4123

"consumedMLUnits": 3.14, # The amount of ML units consumed by the job.

4124

"hyperparameterMetricTag": "A String", # The TensorFlow summary tag name used for optimizing hyperparameter tuning

4125

# trials. See

4126

# [`HyperparameterSpec.hyperparameterMetricTag`](#HyperparameterSpec.FIELDS.hyperparameter_metric_tag)

4127

# for more information. Only set for hyperparameter tuning jobs.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4128

},

4129

"errorMessage": "A String", # Output only. The details of a failure or a cancellation.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4130

"predictionOutput": { # Represents results of a prediction job. # The current prediction job result.

4131

"errorCount": "A String", # The number of data instances which resulted in errors.

4132

"nodeHours": 3.14, # Node hours used by the batch prediction job.

4133

"outputPath": "A String", # The output Google Cloud Storage location provided at the job creation time.

4134

"predictionCount": "A String", # The number of generated predictions.

4135

},

4136

"startTime": "A String", # Output only. When the job processing was started.

4137

"state": "A String", # Output only. The detailed state of a job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4138

"trainingInput": { # Represents input parameters for a training job. When using the gcloud command # Input parameters to create a training job.

4139

# to submit your training job, you can specify the input parameters as

4140

# command-line arguments and/or in a YAML configuration file referenced from

4141

# the --config command-line argument. For details, see the guide to [submitting

4142

# a training job](/ai-platform/training/docs/training-jobs).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4143

"serviceAccount": "A String", # Optional. Specifies the service account for workload run-as account.

4144

# Users submitting jobs must have act-as permission on this run-as account.

4145

# If not specified, then CMLE P4SA will be used by default.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4146

"workerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for workers.

4147

#

4148

# You should only set `workerConfig.acceleratorConfig` if `workerType` is set

4149

# to a Compute Engine machine type. [Learn about restrictions on accelerator

4150

# configurations for

4151

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4152

#

4153

# Set `workerConfig.imageUri` only if you build a custom image for your

4154

# worker. If `workerConfig.imageUri` has not been set, AI Platform uses

4155

# the value of `masterConfig.imageUri`. Learn more about [configuring custom

4156

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4157

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

4158

# [Learn about restrictions on accelerator configurations for

4159

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4160

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

4161

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

4162

# [accelerators for online

4163

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

4164

"type": "A String", # The type of accelerator to use.

4165

"count": "A String", # The number of accelerators to attach to each machine running the job.

4166

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4167

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

4168

# Registry. Learn more about [configuring custom

4169

# containers](/ai-platform/training/docs/distributed-training-containers).

4170

"containerArgs": [ # Arguments to the entrypoint command.

4171

# The following rules apply for container_command and container_args:

4172

# - If you do not supply command or args:

4173

# The defaults defined in the Docker image are used.

4174

# - If you supply a command but no args:

4175

# The default EntryPoint and the default Cmd defined in the Docker image

4176

# are ignored. Your command is run without any arguments.

4177

# - If you supply only args:

4178

# The default Entrypoint defined in the Docker image is run with the args

4179

# that you supplied.

4180

# - If you supply a command and args:

4181

# The default Entrypoint and the default Cmd defined in the Docker image

4182

# are ignored. Your command is run with your args.

4183

# It cannot be set if custom container image is

4184

# not provided.

4185

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4186

# both cannot be set at the same time.

4187

"A String",

4188

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4189

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

4190

# the one used in the custom container. This field is required if the replica

4191

# is a TPU worker that uses a custom container. Otherwise, do not specify

4192

# this field. This must be a [runtime version that currently supports

4193

# training with

4194

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

4195

#

4196

# Note that the version of TensorFlow included in a runtime version may

4197

# differ from the numbering of the runtime version itself, because it may

4198

# have a different [patch

4199

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

4200

# In this field, you must specify the runtime version (TensorFlow minor

4201

# version). For example, if your custom container runs TensorFlow `1.x.y`,

4202

# specify `1.x`.

4203

"containerCommand": [ # The command with which the replica's custom container is run.

4204

# If provided, it will override default ENTRYPOINT of the docker image.

4205

# If not provided, the docker image's ENTRYPOINT is used.

4206

# It cannot be set if custom container image is

4207

# not provided.

4208

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4209

# both cannot be set at the same time.

4210

"A String",

4211

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4212

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4213

"useChiefInTfConfig": True or False, # Optional. Use `chief` instead of `master` in the `TF_CONFIG` environment

4214

# variable when training with a custom container. Defaults to `false`. [Learn

4215

# more about this

4216

# field.](/ai-platform/training/docs/distributed-training-details#chief-versus-master)

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4217

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4218

# This field has no effect for training jobs that don't use a custom

4219

# container.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4220

"masterType": "A String", # Optional. Specifies the type of virtual machine to use for your training

4221

# job's master worker. You must specify this field when `scaleTier` is set to

4222

# `CUSTOM`.

4223

#

4224

# You can use certain Compute Engine machine types directly in this field.

4225

# The following types are supported:

#

# - `n1-standard-4`

# - `n1-standard-8`

# - `n1-standard-16`

# - `n1-standard-32`

# - `n1-standard-64`

# - `n1-standard-96`

# - `n1-highmem-2`

# - `n1-highmem-4`

# - `n1-highmem-8`

# - `n1-highmem-16`

# - `n1-highmem-32`

# - `n1-highmem-64`

# - `n1-highmem-96`

# - `n1-highcpu-16`

# - `n1-highcpu-32`

# - `n1-highcpu-64`

# - `n1-highcpu-96`

#

# Learn more about [using Compute Engine machine

4246

# types](/ml-engine/docs/machine-types#compute-engine-machine-types).

4247

#

4248

# Alternatively, you can use the following legacy machine types:

#

# - `standard`

# - `large_model`

# - `complex_model_s`

4253

# - `complex_model_m`

4254

# - `complex_model_l`

4255

# - `standard_gpu`

4256

# - `complex_model_m_gpu`

4257

# - `complex_model_l_gpu`

4258

# - `standard_p100`

4259

# - `complex_model_m_p100`

4260

# - `standard_v100`

4261

# - `large_model_v100`

4262

# - `complex_model_m_v100`

4263

# - `complex_model_l_v100`

4264

#

4265

# Learn more about [using legacy machine

4266

# types](/ml-engine/docs/machine-types#legacy-machine-types).

4267

#

4268

# Finally, if you want to use a TPU for training, specify `cloud_tpu` in this

4269

# field. Learn more about the [special configuration options for training

4270

# with

4271

# TPUs](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4272

"parameterServerConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for parameter servers.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4273

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4274

# You should only set `parameterServerConfig.acceleratorConfig` if

4275

# `parameterServerType` is set to a Compute Engine machine type. [Learn

4276

# about restrictions on accelerator configurations for

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4277

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4278

#

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4279

# Set `parameterServerConfig.imageUri` only if you build a custom image for

4280

# your parameter server. If `parameterServerConfig.imageUri` has not been

4281

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4282

# containers](/ai-platform/training/docs/distributed-training-containers).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4283

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

4284

# [Learn about restrictions on accelerator configurations for

4285

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4286

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

4287

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

4288

# [accelerators for online

4289

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

4290

"type": "A String", # The type of accelerator to use.

4291

"count": "A String", # The number of accelerators to attach to each machine running the job.

4292

},

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4293

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

4294

# Registry. Learn more about [configuring custom

4295

# containers](/ai-platform/training/docs/distributed-training-containers).

4296

"containerArgs": [ # Arguments to the entrypoint command.

4297

# The following rules apply for container_command and container_args:

4298

# - If you do not supply command or args:

4299

# The defaults defined in the Docker image are used.

4300

# - If you supply a command but no args:

4301

# The default EntryPoint and the default Cmd defined in the Docker image

4302

# are ignored. Your command is run without any arguments.

4303

# - If you supply only args:

4304

# The default Entrypoint defined in the Docker image is run with the args

4305

# that you supplied.

4306

# - If you supply a command and args:

4307

# The default Entrypoint and the default Cmd defined in the Docker image

4308

# are ignored. Your command is run with your args.

4309

# It cannot be set if custom container image is

4310

# not provided.

4311

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4312

# both cannot be set at the same time.

4313

"A String",

4314

],

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4315

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

4316

# the one used in the custom container. This field is required if the replica

4317

# is a TPU worker that uses a custom container. Otherwise, do not specify

4318

# this field. This must be a [runtime version that currently supports

4319

# training with

4320

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

4321

#

4322

# Note that the version of TensorFlow included in a runtime version may

4323

# differ from the numbering of the runtime version itself, because it may

4324

# have a different [patch

4325

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

4326

# In this field, you must specify the runtime version (TensorFlow minor

4327

# version). For example, if your custom container runs TensorFlow `1.x.y`,

4328

# specify `1.x`.

4329

"containerCommand": [ # The command with which the replica's custom container is run.

4330

# If provided, it will override default ENTRYPOINT of the docker image.

4331

# If not provided, the docker image's ENTRYPOINT is used.

4332

# It cannot be set if custom container image is

4333

# not provided.

4334

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4335

# both cannot be set at the same time.

4336

"A String",

4337

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4338

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4339

"region": "A String", # Required. The region to run the training job in. See the [available

4340

# regions](/ai-platform/training/docs/regions) for AI Platform Training.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4341

"jobDir": "A String", # Optional. A Google Cloud Storage path in which to store training outputs

4342

# and other data needed for training. This path is passed to your TensorFlow

4343

# program as the '--job-dir' command-line argument. The benefit of specifying

4344

# this field is that Cloud ML validates the path for use in training.

4345

"pythonVersion": "A String", # Optional. The version of Python used in training. You must either specify

4346

# this field or specify `masterConfig.imageUri`.

4347

#

4348

# The following Python versions are available:

4349

#

4350

# * Python '3.7' is available when `runtime_version` is set to '1.15' or

4351

# later.

4352

# * Python '3.5' is available when `runtime_version` is set to a version

4353

# from '1.4' to '1.14'.

4354

# * Python '2.7' is available when `runtime_version` is set to '1.15' or

4355

# earlier.

4356

#

4357

# Read more about the Python versions available for [each runtime

4358

# version](/ml-engine/docs/runtime-version-list).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4359

"hyperparameters": { # Represents a set of hyperparameters to optimize. # Optional. The set of Hyperparameters to tune.

4360

"maxTrials": 42, # Optional. How many training trials should be attempted to optimize

4361

# the specified hyperparameters.

4362

#

4363

# Defaults to one.

4364

"enableTrialEarlyStopping": True or False, # Optional. Indicates if the hyperparameter tuning job enables auto trial

4365

# early stopping.

4366

"params": [ # Required. The set of parameters to tune.

4367

{ # Represents a single hyperparameter to optimize.

4368

"minValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

4369

# should be unset if type is `CATEGORICAL`. This value should be integers if

4370

# type is INTEGER.

4371

"categoricalValues": [ # Required if type is `CATEGORICAL`. The list of possible categories.

4372

"A String",

4373

],

4374

"scaleType": "A String", # Optional. How the parameter should be scaled to the hypercube.

4375

# Leave unset for categorical parameters.

4376

# Some kind of scaling is strongly recommended for real or integral

4377

# parameters (e.g., `UNIT_LINEAR_SCALE`).

4378

"discreteValues": [ # Required if type is `DISCRETE`.

4379

# A list of feasible points.

4380

# The list should be in strictly increasing order. For instance, this

4381

# parameter might have possible settings of 1.5, 2.5, and 4.0. This list

4382

# should not contain more than 1,000 values.

4383

3.14,

4384

],

4385

"type": "A String", # Required. The type of the parameter.

4386

"maxValue": 3.14, # Required if type is `DOUBLE` or `INTEGER`. This field

4387

# should be unset if type is `CATEGORICAL`. This value should be integers if

4388

# type is `INTEGER`.

4389

"parameterName": "A String", # Required. The parameter name must be unique amongst all ParameterConfigs in

4390

# a HyperparameterSpec message. E.g., "learning_rate".

4391

},

4392

],

4393

"maxFailedTrials": 42, # Optional. The number of failed trials that need to be seen before failing

4394

# the hyperparameter tuning job. You can specify this field to override the

4395

# default failing criteria for AI Platform hyperparameter tuning jobs.

4396

#

4397

# Defaults to zero, which means the service decides when a hyperparameter

4398

# job should fail.

4399

"hyperparameterMetricTag": "A String", # Optional. The TensorFlow summary tag name to use for optimizing trials. For

4400

# current versions of TensorFlow, this tag name should exactly match what is

4401

# shown in TensorBoard, including all scopes. For versions of TensorFlow

4402

# prior to 0.12, this should be only the tag passed to tf.Summary.

4403

# By default, "training/hptuning/metric" will be used.

4404

"resumePreviousJobId": "A String", # Optional. The prior hyperparameter tuning job id that users hope to

4405

# continue with. The job id will be used to find the corresponding vizier

4406

# study guid and resume the study.

4407

"goal": "A String", # Required. The type of goal to use for tuning. Available types are

4408

# `MAXIMIZE` and `MINIMIZE`.

4409

#

4410

# Defaults to `MAXIMIZE`.

4411

"algorithm": "A String", # Optional. The search algorithm specified for the hyperparameter

4412

# tuning job.

4413

# Uses the default AI Platform hyperparameter tuning

4414

# algorithm if unspecified.

4415

"maxParallelTrials": 42, # Optional. The number of training trials to run concurrently.

4416

# You can reduce the time it takes to perform hyperparameter tuning by adding

4417

# trials in parallel. However, each trail only benefits from the information

4418

# gained in completed trials. That means that a trial does not get access to

4419

# the results of trials running at the same time, which could reduce the

4420

# quality of the overall optimization.

4421

#

4422

# Each trial will use the same scale tier and machine types.

#

# Defaults to one.

},

"evaluatorType": "A String", # Optional. Specifies the type of virtual machine to use for your training

4427

# job's evaluator nodes.

4428

#

4429

# The supported values are the same as those described in the entry for

4430

# `masterType`.

4431

#

4432

# This value must be consistent with the category of machine type that

4433

# `masterType` uses. In other words, both must be Compute Engine machine

4434

# types or both must be legacy machine types.

4435

#

4436

# This value must be present when `scaleTier` is set to `CUSTOM` and

4437

# `evaluatorCount` is greater than zero.

4438

"network": "A String", # Optional. The full name of the Google Compute Engine

4439

# [network](/compute/docs/networks-and-firewalls#networks) to which the Job

4440

# is peered. For example, projects/12345/global/networks/myVPC. Format is of

4441

# the form projects/{project}/global/networks/{network}. Where {project} is a

4442

# project number, as in '12345', and {network} is network name.".

4443

#

4444

# Private services access must already be configured for the network. If left

4445

# unspecified, the Job is not peered with any network. Learn more -

4446

# Connecting Job to user network over private

4447

# IP.

4448

"parameterServerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

4449

# job's parameter server.

4450

#

4451

# The supported values are the same as those described in the entry for

4452

# `master_type`.

4453

#

4454

# This value must be consistent with the category of machine type that

4455

# `masterType` uses. In other words, both must be Compute Engine machine

4456

# types or both must be legacy machine types.

4457

#

4458

# This value must be present when `scaleTier` is set to `CUSTOM` and

4459

# `parameter_server_count` is greater than zero.

4460

"workerType": "A String", # Optional. Specifies the type of virtual machine to use for your training

4461

# job's worker nodes.

4462

#

4463

# The supported values are the same as those described in the entry for

4464

# `masterType`.

4465

#

4466

# This value must be consistent with the category of machine type that

4467

# `masterType` uses. In other words, both must be Compute Engine machine

4468

# types or both must be legacy machine types.

4469

#

4470

# If you use `cloud_tpu` for this value, see special instructions for

4471

# [configuring a custom TPU

4472

# machine](/ml-engine/docs/tensorflow/using-tpus#configuring_a_custom_tpu_machine).

4473

#

4474

# This value must be present when `scaleTier` is set to `CUSTOM` and

4475

# `workerCount` is greater than zero.

4476

"masterConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for your master worker.

4477

#

4478

# You should only set `masterConfig.acceleratorConfig` if `masterType` is set

4479

# to a Compute Engine machine type. Learn about [restrictions on accelerator

4480

# configurations for

4481

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4482

#

4483

# Set `masterConfig.imageUri` only if you build a custom image. Only one of

4484

# `masterConfig.imageUri` and `runtimeVersion` should be set. Learn more

4485

# about [configuring custom

4486

# containers](/ai-platform/training/docs/distributed-training-containers).

4487

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

4488

# [Learn about restrictions on accelerator configurations for

4489

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4490

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

4491

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

4492

# [accelerators for online

4493

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

4494

"type": "A String", # The type of accelerator to use.

4495

"count": "A String", # The number of accelerators to attach to each machine running the job.

4496

},

4497

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

4498

# Registry. Learn more about [configuring custom

4499

# containers](/ai-platform/training/docs/distributed-training-containers).

4500

"containerArgs": [ # Arguments to the entrypoint command.

4501

# The following rules apply for container_command and container_args:

4502

# - If you do not supply command or args:

4503

# The defaults defined in the Docker image are used.

4504

# - If you supply a command but no args:

4505

# The default EntryPoint and the default Cmd defined in the Docker image

4506

# are ignored. Your command is run without any arguments.

4507

# - If you supply only args:

4508

# The default Entrypoint defined in the Docker image is run with the args

4509

# that you supplied.

4510

# - If you supply a command and args:

4511

# The default Entrypoint and the default Cmd defined in the Docker image

4512

# are ignored. Your command is run with your args.

4513

# It cannot be set if custom container image is

4514

# not provided.

4515

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4516

# both cannot be set at the same time.

4517

"A String",

4518

],

4519

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

4520

# the one used in the custom container. This field is required if the replica

4521

# is a TPU worker that uses a custom container. Otherwise, do not specify

4522

# this field. This must be a [runtime version that currently supports

4523

# training with

4524

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

4525

#

4526

# Note that the version of TensorFlow included in a runtime version may

4527

# differ from the numbering of the runtime version itself, because it may

4528

# have a different [patch

4529

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

4530

# In this field, you must specify the runtime version (TensorFlow minor

4531

# version). For example, if your custom container runs TensorFlow `1.x.y`,

4532

# specify `1.x`.

4533

"containerCommand": [ # The command with which the replica's custom container is run.

4534

# If provided, it will override default ENTRYPOINT of the docker image.

4535

# If not provided, the docker image's ENTRYPOINT is used.

4536

# It cannot be set if custom container image is

4537

# not provided.

4538

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4539

# both cannot be set at the same time.

"A String",

],

},

"evaluatorCount": "A String", # Optional. The number of evaluator replicas to use for the training job.

4544

# Each replica in the cluster will be of the type specified in

4545

# `evaluator_type`.

4546

#

4547

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

4548

# set this value, you must also set `evaluator_type`.

4549

#

4550

# The default value is zero.

4551

"args": [ # Optional. Command-line arguments passed to the training application when it

4552

# starts. If your job uses a custom container, then the arguments are passed

4553

# to the container's <a class="external" target="_blank"

4554

# href="https://docs.docker.com/engine/reference/builder/#entrypoint">

4555

# `ENTRYPOINT`</a> command.

4556

"A String",

4557

],

4558

"pythonModule": "A String", # Required. The Python module name to run after installing the packages.

4559

"runtimeVersion": "A String", # Optional. The AI Platform runtime version to use for training. You must

4560

# either specify this field or specify `masterConfig.imageUri`.

4561

#

4562

# For more information, see the [runtime version

4563

# list](/ai-platform/training/docs/runtime-version-list) and learn [how to

4564

# manage runtime versions](/ai-platform/training/docs/versioning).

4565

"parameterServerCount": "A String", # Optional. The number of parameter server replicas to use for the training

4566

# job. Each replica in the cluster will be of the type specified in

4567

# `parameter_server_type`.

4568

#

4569

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

4570

# set this value, you must also set `parameter_server_type`.

4571

#

4572

# The default value is zero.

4573

"evaluatorConfig": { # Represents the configuration for a replica in a cluster. # Optional. The configuration for evaluators.

4574

#

4575

# You should only set `evaluatorConfig.acceleratorConfig` if

4576

# `evaluatorType` is set to a Compute Engine machine type. [Learn

4577

# about restrictions on accelerator configurations for

4578

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4579

#

4580

# Set `evaluatorConfig.imageUri` only if you build a custom image for

4581

# your evaluator. If `evaluatorConfig.imageUri` has not been

4582

# set, AI Platform uses the value of `masterConfig.imageUri`. Learn more about [configuring custom

4583

# containers](/ai-platform/training/docs/distributed-training-containers).

4584

"acceleratorConfig": { # Represents a hardware accelerator request config. # Represents the type and number of accelerators used by the replica.

4585

# [Learn about restrictions on accelerator configurations for

4586

# training.](/ai-platform/training/docs/using-gpus#compute-engine-machine-types-with-gpu)

4587

# Note that the AcceleratorConfig can be used in both Jobs and Versions.

4588

# Learn more about [accelerators for training](/ml-engine/docs/using-gpus) and

4589

# [accelerators for online

4590

# prediction](/ml-engine/docs/machine-types-online-prediction#gpus).

4591

"type": "A String", # The type of accelerator to use.

4592

"count": "A String", # The number of accelerators to attach to each machine running the job.

4593

},

4594

"imageUri": "A String", # The Docker image to run on the replica. This image must be in Container

4595

# Registry. Learn more about [configuring custom

4596

# containers](/ai-platform/training/docs/distributed-training-containers).

4597

"containerArgs": [ # Arguments to the entrypoint command.

4598

# The following rules apply for container_command and container_args:

4599

# - If you do not supply command or args:

4600

# The defaults defined in the Docker image are used.

4601

# - If you supply a command but no args:

4602

# The default EntryPoint and the default Cmd defined in the Docker image

4603

# are ignored. Your command is run without any arguments.

4604

# - If you supply only args:

4605

# The default Entrypoint defined in the Docker image is run with the args

4606

# that you supplied.

4607

# - If you supply a command and args:

4608

# The default Entrypoint and the default Cmd defined in the Docker image

4609

# are ignored. Your command is run with your args.

4610

# It cannot be set if custom container image is

4611

# not provided.

4612

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4613

# both cannot be set at the same time.

4614

"A String",

4615

],

4616

"tpuTfVersion": "A String", # The AI Platform runtime version that includes a TensorFlow version matching

4617

# the one used in the custom container. This field is required if the replica

4618

# is a TPU worker that uses a custom container. Otherwise, do not specify

4619

# this field. This must be a [runtime version that currently supports

4620

# training with

4621

# TPUs](/ml-engine/docs/tensorflow/runtime-version-list#tpu-support).

4622

#

4623

# Note that the version of TensorFlow included in a runtime version may

4624

# differ from the numbering of the runtime version itself, because it may

4625

# have a different [patch

4626

# version](https://www.tensorflow.org/guide/version_compat#semantic_versioning_20).

4627

# In this field, you must specify the runtime version (TensorFlow minor

4628

# version). For example, if your custom container runs TensorFlow `1.x.y`,

4629

# specify `1.x`.

4630

"containerCommand": [ # The command with which the replica's custom container is run.

4631

# If provided, it will override default ENTRYPOINT of the docker image.

4632

# If not provided, the docker image's ENTRYPOINT is used.

4633

# It cannot be set if custom container image is

4634

# not provided.

4635

# Note that this field and [TrainingInput.args] are mutually exclusive, i.e.,

4636

# both cannot be set at the same time.

"A String",

],

},

"encryptionConfig": { # Represents a custom encryption key configuration that can be applied to # Optional. Options for using customer-managed encryption keys (CMEK) to

4641

# protect resources created by a training job, instead of using Google's

4642

# default encryption. If this is set, then all resources created by the

4643

# training job will be encrypted with the customer-managed encryption key

4644

# that you specify.

4645

#

4646

# [Learn how and when to use CMEK with AI Platform

4647

# Training](/ai-platform/training/docs/cmek).

4648

# a resource.

4649

"kmsKeyName": "A String", # The Cloud KMS resource identifier of the customer-managed encryption key

4650

# used to protect a resource, such as a training job. It has the following

4651

# format:

4652

# `projects/{PROJECT_ID}/locations/{REGION}/keyRings/{KEY_RING_NAME}/cryptoKeys/{KEY_NAME}`

4653

},

4654

"workerCount": "A String", # Optional. The number of worker replicas to use for the training job. Each

4655

# replica in the cluster will be of the type specified in `worker_type`.

4656

#

4657

# This value can only be used when `scale_tier` is set to `CUSTOM`. If you

4658

# set this value, you must also set `worker_type`.

4659

#

4660

# The default value is zero.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4661

"scheduling": { # All parameters related to scheduling of training jobs. # Optional. Scheduling options for a training job.

4662

"maxWaitTime": "A String",

4663

"maxRunningTime": "A String", # Optional. The maximum job running time, expressed in seconds. The field can

4664

# contain up to nine fractional digits, terminated by `s`. If not specified,

4665

# this field defaults to `604800s` (seven days).

4666

#

4667

# If the training job is still running after this duration, AI Platform

4668

# Training cancels it.

4669

#

4670

# For example, if you want to ensure your job runs for no more than 2 hours,

4671

# set this field to `7200s` (2 hours * 60 minutes / hour * 60 seconds /

4672

# minute).

4673

#

4674

# If you submit your training job using the `gcloud` tool, you can [provide

4675

# this field in a `config.yaml`

4676

# file](/ai-platform/training/docs/training-jobs#formatting_your_configuration_parameters).

# For example:

#

# ```yaml

# trainingInput:

# ...

# scheduling:

# maxRunningTime: 7200s

4684

# ...

4685

# ```

4686

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4687

"scaleTier": "A String", # Required. Specifies the machine types, the number of replicas for workers

4688

# and parameter servers.

4689

"packageUris": [ # Required. The Google Cloud Storage location of the packages with

4690

# the training program and any additional dependencies.

4691

# The maximum number of package URIs is 100.

4692

"A String",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4693

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4694

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4695

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

4696

# prevent simultaneous updates of a job from overwriting each other.

4697

# It is strongly suggested that systems make use of the `etag` in the

4698

# read-modify-write cycle to perform job updates in order to avoid race

4699

# conditions: An `etag` is returned in the response to `GetJob`, and

4700

# systems are expected to put that etag in the request to `UpdateJob` to

4701

# ensure that their change will be applied to the same version of the job.

4702

"jobId": "A String", # Required. The user-specified id of the job.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4703

}</pre>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

</div>

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4707

<code class="details" id="setIamPolicy">setIamPolicy(resource, body=None, x__xgafv=None)</code>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4708

<pre>Sets the access control policy on the specified resource. Replaces any

4709

existing policy.

4710

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4711

Can return `NOT_FOUND`, `INVALID_ARGUMENT`, and `PERMISSION_DENIED` errors.

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4712

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4713

Args:

4714

resource: string, REQUIRED: The resource for which the policy is being specified.

4715

See the operation documentation for the appropriate value for this field. (required)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4716

body: object, The request body.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4717

The object takes the form of:

4718

4719

{ # Request message for `SetIamPolicy` method.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4720

"policy": { # An Identity and Access Management (IAM) policy, which specifies access # REQUIRED: The complete policy to be applied to the `resource`. The size of

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4721

# the policy is limited to a few 10s of KB. An empty policy is a

4722

# valid policy but certain Cloud Platform services (such as Projects)

4723

# might reject them.

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4724

# controls for Google Cloud resources.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4725

#

4726

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4727

# A `Policy` is a collection of `bindings`. A `binding` binds one or more

4728

# `members` to a single `role`. Members can be user accounts, service accounts,

4729

# Google groups, and domains (such as G Suite). A `role` is a named list of

4730

# permissions; each `role` can be an IAM predefined role or a user-created

4731

# custom role.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4732

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4733

# For some types of Google Cloud resources, a `binding` can also specify a

4734

# `condition`, which is a logical expression that allows access to a resource

4735

# only if the expression evaluates to `true`. A condition can add constraints

4736

# based on attributes of the request, the resource, or both. To learn which

4737

# resources support conditions in their IAM policies, see the

4738

# [IAM documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4739

#

4740

# **JSON example:**

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4741

#

4742

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4743

# "bindings": [

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4744

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4745

# "role": "roles/resourcemanager.organizationAdmin",

4746

# "members": [

4747

# "user:mike@example.com",

4748

# "group:admins@example.com",

4749

# "domain:google.com",

4750

# "serviceAccount:my-project-id@appspot.gserviceaccount.com"

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4751

# ]

4752

# },

4753

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4754

# "role": "roles/resourcemanager.organizationViewer",

4755

# "members": [

4756

# "user:eve@example.com"

4757

# ],

4758

# "condition": {

4759

# "title": "expirable access",

4760

# "description": "Does not grant access after Sep 2020",

4761

# "expression": "request.time < timestamp('2020-10-01T00:00:00.000Z')",

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4762

# }

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4763

# }

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4764

# ],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4765

# "etag": "BwWWja0YfJA=",

4766

# "version": 3

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4767

# }

4768

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4769

# **YAML example:**

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

#

# bindings:

# - members:

# - user:mike@example.com

4774

# - group:admins@example.com

4775

# - domain:google.com

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4776

# - serviceAccount:my-project-id@appspot.gserviceaccount.com

4777

# role: roles/resourcemanager.organizationAdmin

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4778

# - members:

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4779

# - user:eve@example.com

4780

# role: roles/resourcemanager.organizationViewer

4781

# condition:

4782

# title: expirable access

4783

# description: Does not grant access after Sep 2020

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4784

# expression: request.time < timestamp('2020-10-01T00:00:00.000Z')

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4785

# - etag: BwWWja0YfJA=

4786

# - version: 3

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4787

#

4788

# For a description of IAM and its features, see the

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4789

# [IAM documentation](https://cloud.google.com/iam/docs/).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4790

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

4791

# prevent simultaneous updates of a policy from overwriting each other.

4792

# It is strongly suggested that systems make use of the `etag` in the

4793

# read-modify-write cycle to perform policy updates in order to avoid race

4794

# conditions: An `etag` is returned in the response to `getIamPolicy`, and

4795

# systems are expected to put that etag in the request to `setIamPolicy` to

4796

# ensure that their change will be applied to the same version of the policy.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4797

#

4798

# **Important:** If you use IAM Conditions, you must include the `etag` field

4799

# whenever you call `setIamPolicy`. If you omit this field, then IAM allows

4800

# you to overwrite a version `3` policy with a version `1` policy, and all of

4801

# the conditions in the version `3` policy are lost.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4802

"auditConfigs": [ # Specifies cloud audit logging configuration for this policy.

4803

{ # Specifies the audit configuration for a service.

4804

# The configuration determines which permission types are logged, and what

4805

# identities, if any, are exempted from logging.

4806

# An AuditConfig must have one or more AuditLogConfigs.

4807

#

4808

# If there are AuditConfigs for both `allServices` and a specific service,

4809

# the union of the two AuditConfigs is used for that service: the log_types

4810

# specified in each AuditConfig are enabled, and the exempted_members in each

4811

# AuditLogConfig are exempted.

4812

#

4813

# Example Policy with multiple AuditConfigs:

#

# {

# "audit_configs": [

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4818

# "service": "allServices",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4819

# "audit_log_configs": [

4820

# {

4821

# "log_type": "DATA_READ",

4822

# "exempted_members": [

4823

# "user:jose@example.com"

4824

# ]

4825

# },

4826

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4827

# "log_type": "DATA_WRITE"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4828

# },

4829

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4830

# "log_type": "ADMIN_READ"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

# }

# ]

# },

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4835

# "service": "sampleservice.googleapis.com",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4836

# "audit_log_configs": [

4837

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4838

# "log_type": "DATA_READ"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4839

# },

4840

# {

4841

# "log_type": "DATA_WRITE",

4842

# "exempted_members": [

4843

# "user:aliya@example.com"

# ]

# }

# ]

# }

# ]

# }

#

# For sampleservice, this policy enables DATA_READ, DATA_WRITE and ADMIN_READ

4852

# logging. It also exempts jose@example.com from DATA_READ logging, and

4853

# aliya@example.com from DATA_WRITE logging.

4854

"service": "A String", # Specifies a service that will be enabled for audit logging.

4855

# For example, `storage.googleapis.com`, `cloudsql.googleapis.com`.

4856

# `allServices` is a special value that covers all services.

4857

"auditLogConfigs": [ # The configuration for logging of each type of permission.

4858

{ # Provides the configuration for logging a type of permissions.

# Example:

#

# {

# "audit_log_configs": [

4863

# {

4864

# "log_type": "DATA_READ",

4865

# "exempted_members": [

4866

# "user:jose@example.com"

4867

# ]

4868

# },

4869

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4870

# "log_type": "DATA_WRITE"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

# }

# ]

# }

#

# This enables 'DATA_READ' and 'DATA_WRITE' logging, while exempting

4876

# jose@example.com from DATA_READ logging.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4877

"logType": "A String", # The log type that this config enables.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4878

"exemptedMembers": [ # Specifies the identities that do not cause logging for this type of

4879

# permission.

4880

# Follows the same format of Binding.members.

4881

"A String",

4882

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

},

],

},

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4887

"version": 42, # Specifies the format of the policy.

4888

#

4889

# Valid values are `0`, `1`, and `3`. Requests that specify an invalid value

4890

# are rejected.

4891

#

4892

# Any operation that affects conditional role bindings must specify version

4893

# `3`. This requirement applies to the following operations:

4894

#

4895

# * Getting a policy that includes a conditional role binding

4896

# * Adding a conditional role binding to a policy

4897

# * Changing a conditional role binding in a policy

4898

# * Removing any role binding, with or without a condition, from a policy

4899

# that includes conditions

4900

#

4901

# **Important:** If you use IAM Conditions, you must include the `etag` field

4902

# whenever you call `setIamPolicy`. If you omit this field, then IAM allows

4903

# you to overwrite a version `3` policy with a version `1` policy, and all of

4904

# the conditions in the version `3` policy are lost.

4905

#

4906

# If a policy does not include any conditions, operations on that policy may

4907

# specify any valid version or leave the field unset.

4908

#

4909

# To learn which resources support conditions in their IAM policies, see the

4910

# [IAM documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4911

"bindings": [ # Associates a list of `members` to a `role`. Optionally, may specify a

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4912

# `condition` that determines how and when the `bindings` are applied. Each

4913

# of the `bindings` must contain at least one member.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4914

{ # Associates `members` with a `role`.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4915

"role": "A String", # Role that is assigned to `members`.

4916

# For example, `roles/viewer`, `roles/editor`, or `roles/owner`.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4917

"condition": { # Represents a textual expression in the Common Expression Language (CEL) # The condition that is associated with this binding.

4918

#

4919

# If the condition evaluates to `true`, then this binding applies to the

4920

# current request.

4921

#

4922

# If the condition evaluates to `false`, then this binding does not apply to

4923

# the current request. However, a different role binding might grant the same

4924

# role to one or more of the members in this binding.

4925

#

4926

# To learn which resources support conditions in their IAM policies, see the

4927

# [IAM

4928

# documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

4929

# syntax. CEL is a C-like expression language. The syntax and semantics of CEL

4930

# are documented at https://github.com/google/cel-spec.

4931

#

4932

# Example (Comparison):

4933

#

4934

# title: "Summary size limit"

4935

# description: "Determines if a summary is less than 100 chars"

4936

# expression: "document.summary.size() < 100"

4937

#

4938

# Example (Equality):

4939

#

4940

# title: "Requestor is owner"

4941

# description: "Determines if requestor is the document owner"

4942

# expression: "document.owner == request.auth.claims.email"

#

# Example (Logic):

#

# title: "Public documents"

4947

# description: "Determine whether the document should be publicly visible"

4948

# expression: "document.type != 'private' && document.type != 'internal'"

4949

#

4950

# Example (Data Manipulation):

4951

#

4952

# title: "Notification string"

4953

# description: "Create a notification string with a timestamp."

4954

# expression: "'New message received at ' + string(document.create_time)"

4955

#

4956

# The exact variables and functions that may be referenced within an expression

4957

# are determined by the service that evaluates it. See the service

4958

# documentation for additional information.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

4959

"expression": "A String", # Textual representation of an expression in Common Expression Language

4960

# syntax.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4961

"title": "A String", # Optional. Title for the expression, i.e. a short string describing

4962

# its purpose. This can be used e.g. in UIs which allow to enter the

4963

# expression.

4964

"location": "A String", # Optional. String indicating the location of the expression for error

4965

# reporting, e.g. a file name and a position in the file.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

4966

"description": "A String", # Optional. Description of the expression. This is a longer text which

4967

# describes the expression, e.g. when hovered over it in a UI.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

4968

},

4969

"members": [ # Specifies the identities requesting access for a Cloud Platform resource.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4970

# `members` can have the following values:

4971

#

4972

# * `allUsers`: A special identifier that represents anyone who is

4973

# on the internet; with or without a Google account.

4974

#

4975

# * `allAuthenticatedUsers`: A special identifier that represents anyone

4976

# who is authenticated with a Google account or a service account.

4977

#

4978

# * `user:{emailid}`: An email address that represents a specific Google

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4979

# account. For example, `alice@example.com` .

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

4980

#

4981

#

4982

# * `serviceAccount:{emailid}`: An email address that represents a service

4983

# account. For example, `my-other-app@appspot.gserviceaccount.com`.

4984

#

4985

# * `group:{emailid}`: An email address that represents a Google group.

4986

# For example, `admins@example.com`.

4987

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

4988

# * `deleted:user:{emailid}?uid={uniqueid}`: An email address (plus unique

4989

# identifier) representing a user that has been recently deleted. For

4990

# example, `alice@example.com?uid=123456789012345678901`. If the user is

4991

# recovered, this value reverts to `user:{emailid}` and the recovered user

4992

# retains the role in the binding.

4993

#

4994

# * `deleted:serviceAccount:{emailid}?uid={uniqueid}`: An email address (plus

4995

# unique identifier) representing a service account that has been recently

4996

# deleted. For example,

4997

# `my-other-app@appspot.gserviceaccount.com?uid=123456789012345678901`.

4998

# If the service account is undeleted, this value reverts to

4999

# `serviceAccount:{emailid}` and the undeleted service account retains the

5000

# role in the binding.

5001

#

5002

# * `deleted:group:{emailid}?uid={uniqueid}`: An email address (plus unique

5003

# identifier) representing a Google group that has been recently

5004

# deleted. For example, `admins@example.com?uid=123456789012345678901`. If

5005

# the group is recovered, this value reverts to `group:{emailid}` and the

5006

# recovered group retains the role in the binding.

5007

#

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5008

#

5009

# * `domain:{domain}`: The G Suite domain (primary) that represents all the

5010

# users of that domain. For example, `google.com` or `example.com`.

5011

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5012

"A String",

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5013

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5014

},

5015

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5016

},

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5017

"updateMask": "A String", # OPTIONAL: A FieldMask specifying which fields of the policy to modify. Only

5018

# the fields in the mask will be modified. If no mask is provided, the

5019

# following default mask is used:

5020

#

5021

# `paths: "bindings, etag"`

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5022

}

5023

5024

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

5031

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5032

{ # An Identity and Access Management (IAM) policy, which specifies access

5033

# controls for Google Cloud resources.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5034

#

5035

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5036

# A `Policy` is a collection of `bindings`. A `binding` binds one or more

5037

# `members` to a single `role`. Members can be user accounts, service accounts,

5038

# Google groups, and domains (such as G Suite). A `role` is a named list of

5039

# permissions; each `role` can be an IAM predefined role or a user-created

5040

# custom role.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5041

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5042

# For some types of Google Cloud resources, a `binding` can also specify a

5043

# `condition`, which is a logical expression that allows access to a resource

5044

# only if the expression evaluates to `true`. A condition can add constraints

5045

# based on attributes of the request, the resource, or both. To learn which

5046

# resources support conditions in their IAM policies, see the

5047

# [IAM documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5048

#

5049

# **JSON example:**

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5050

#

5051

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5052

# "bindings": [

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5053

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5054

# "role": "roles/resourcemanager.organizationAdmin",

5055

# "members": [

5056

# "user:mike@example.com",

5057

# "group:admins@example.com",

5058

# "domain:google.com",

5059

# "serviceAccount:my-project-id@appspot.gserviceaccount.com"

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5060

# ]

5061

# },

5062

# {

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5063

# "role": "roles/resourcemanager.organizationViewer",

5064

# "members": [

5065

# "user:eve@example.com"

5066

# ],

5067

# "condition": {

5068

# "title": "expirable access",

5069

# "description": "Does not grant access after Sep 2020",

5070

# "expression": "request.time < timestamp('2020-10-01T00:00:00.000Z')",

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5071

# }

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5072

# }

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5073

# ],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5074

# "etag": "BwWWja0YfJA=",

5075

# "version": 3

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5076

# }

5077

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5078

# **YAML example:**

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

#

# bindings:

# - members:

# - user:mike@example.com

5083

# - group:admins@example.com

5084

# - domain:google.com

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5085

# - serviceAccount:my-project-id@appspot.gserviceaccount.com

5086

# role: roles/resourcemanager.organizationAdmin

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5087

# - members:

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5088

# - user:eve@example.com

5089

# role: roles/resourcemanager.organizationViewer

5090

# condition:

5091

# title: expirable access

5092

# description: Does not grant access after Sep 2020

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5093

# expression: request.time < timestamp('2020-10-01T00:00:00.000Z')

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5094

# - etag: BwWWja0YfJA=

5095

# - version: 3

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5096

#

5097

# For a description of IAM and its features, see the

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5098

# [IAM documentation](https://cloud.google.com/iam/docs/).

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5099

"etag": "A String", # `etag` is used for optimistic concurrency control as a way to help

5100

# prevent simultaneous updates of a policy from overwriting each other.

5101

# It is strongly suggested that systems make use of the `etag` in the

5102

# read-modify-write cycle to perform policy updates in order to avoid race

5103

# conditions: An `etag` is returned in the response to `getIamPolicy`, and

5104

# systems are expected to put that etag in the request to `setIamPolicy` to

5105

# ensure that their change will be applied to the same version of the policy.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5106

#

5107

# **Important:** If you use IAM Conditions, you must include the `etag` field

5108

# whenever you call `setIamPolicy`. If you omit this field, then IAM allows

5109

# you to overwrite a version `3` policy with a version `1` policy, and all of

5110

# the conditions in the version `3` policy are lost.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5111

"auditConfigs": [ # Specifies cloud audit logging configuration for this policy.

5112

{ # Specifies the audit configuration for a service.

5113

# The configuration determines which permission types are logged, and what

5114

# identities, if any, are exempted from logging.

5115

# An AuditConfig must have one or more AuditLogConfigs.

5116

#

5117

# If there are AuditConfigs for both `allServices` and a specific service,

5118

# the union of the two AuditConfigs is used for that service: the log_types

5119

# specified in each AuditConfig are enabled, and the exempted_members in each

5120

# AuditLogConfig are exempted.

5121

#

5122

# Example Policy with multiple AuditConfigs:

#

# {

# "audit_configs": [

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5127

# "service": "allServices",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5128

# "audit_log_configs": [

5129

# {

5130

# "log_type": "DATA_READ",

5131

# "exempted_members": [

5132

# "user:jose@example.com"

5133

# ]

5134

# },

5135

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5136

# "log_type": "DATA_WRITE"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5137

# },

5138

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5139

# "log_type": "ADMIN_READ"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

# }

# ]

# },

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5144

# "service": "sampleservice.googleapis.com",

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5145

# "audit_log_configs": [

5146

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5147

# "log_type": "DATA_READ"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5148

# },

5149

# {

5150

# "log_type": "DATA_WRITE",

5151

# "exempted_members": [

5152

# "user:aliya@example.com"

# ]

# }

# ]

# }

# ]

# }

#

# For sampleservice, this policy enables DATA_READ, DATA_WRITE and ADMIN_READ

5161

# logging. It also exempts jose@example.com from DATA_READ logging, and

5162

# aliya@example.com from DATA_WRITE logging.

5163

"service": "A String", # Specifies a service that will be enabled for audit logging.

5164

# For example, `storage.googleapis.com`, `cloudsql.googleapis.com`.

5165

# `allServices` is a special value that covers all services.

5166

"auditLogConfigs": [ # The configuration for logging of each type of permission.

5167

{ # Provides the configuration for logging a type of permissions.

# Example:

#

# {

# "audit_log_configs": [

5172

# {

5173

# "log_type": "DATA_READ",

5174

# "exempted_members": [

5175

# "user:jose@example.com"

5176

# ]

5177

# },

5178

# {

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5179

# "log_type": "DATA_WRITE"

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

# }

# ]

# }

#

# This enables 'DATA_READ' and 'DATA_WRITE' logging, while exempting

5185

# jose@example.com from DATA_READ logging.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5186

"logType": "A String", # The log type that this config enables.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5187

"exemptedMembers": [ # Specifies the identities that do not cause logging for this type of

5188

# permission.

5189

# Follows the same format of Binding.members.

5190

"A String",

5191

],

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

},

],

},

],

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5196

"version": 42, # Specifies the format of the policy.

5197

#

5198

# Valid values are `0`, `1`, and `3`. Requests that specify an invalid value

5199

# are rejected.

5200

#

5201

# Any operation that affects conditional role bindings must specify version

5202

# `3`. This requirement applies to the following operations:

5203

#

5204

# * Getting a policy that includes a conditional role binding

5205

# * Adding a conditional role binding to a policy

5206

# * Changing a conditional role binding in a policy

5207

# * Removing any role binding, with or without a condition, from a policy

5208

# that includes conditions

5209

#

5210

# **Important:** If you use IAM Conditions, you must include the `etag` field

5211

# whenever you call `setIamPolicy`. If you omit this field, then IAM allows

5212

# you to overwrite a version `3` policy with a version `1` policy, and all of

5213

# the conditions in the version `3` policy are lost.

5214

#

5215

# If a policy does not include any conditions, operations on that policy may

5216

# specify any valid version or leave the field unset.

5217

#

5218

# To learn which resources support conditions in their IAM policies, see the

5219

# [IAM documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5220

"bindings": [ # Associates a list of `members` to a `role`. Optionally, may specify a

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5221

# `condition` that determines how and when the `bindings` are applied. Each

5222

# of the `bindings` must contain at least one member.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5223

{ # Associates `members` with a `role`.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5224

"role": "A String", # Role that is assigned to `members`.

5225

# For example, `roles/viewer`, `roles/editor`, or `roles/owner`.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5226

"condition": { # Represents a textual expression in the Common Expression Language (CEL) # The condition that is associated with this binding.

5227

#

5228

# If the condition evaluates to `true`, then this binding applies to the

5229

# current request.

5230

#

5231

# If the condition evaluates to `false`, then this binding does not apply to

5232

# the current request. However, a different role binding might grant the same

5233

# role to one or more of the members in this binding.

5234

#

5235

# To learn which resources support conditions in their IAM policies, see the

5236

# [IAM

5237

# documentation](https://cloud.google.com/iam/help/conditions/resource-policies).

5238

# syntax. CEL is a C-like expression language. The syntax and semantics of CEL

5239

# are documented at https://github.com/google/cel-spec.

5240

#

5241

# Example (Comparison):

5242

#

5243

# title: "Summary size limit"

5244

# description: "Determines if a summary is less than 100 chars"

5245

# expression: "document.summary.size() < 100"

5246

#

5247

# Example (Equality):

5248

#

5249

# title: "Requestor is owner"

5250

# description: "Determines if requestor is the document owner"

5251

# expression: "document.owner == request.auth.claims.email"

#

# Example (Logic):

#

# title: "Public documents"

5256

# description: "Determine whether the document should be publicly visible"

5257

# expression: "document.type != 'private' && document.type != 'internal'"

5258

#

5259

# Example (Data Manipulation):

5260

#

5261

# title: "Notification string"

5262

# description: "Create a notification string with a timestamp."

5263

# expression: "'New message received at ' + string(document.create_time)"

5264

#

5265

# The exact variables and functions that may be referenced within an expression

5266

# are determined by the service that evaluates it. See the service

5267

# documentation for additional information.

Bu Sun Kim

2020-05-27 12:20:54 -0700

[diff] [blame]

5268

"expression": "A String", # Textual representation of an expression in Common Expression Language

5269

# syntax.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5270

"title": "A String", # Optional. Title for the expression, i.e. a short string describing

5271

# its purpose. This can be used e.g. in UIs which allow to enter the

5272

# expression.

5273

"location": "A String", # Optional. String indicating the location of the expression for error

5274

# reporting, e.g. a file name and a position in the file.

Bu Sun Kim

2020-07-22 17:02:09 -0700

[diff] [blame]

5275

"description": "A String", # Optional. Description of the expression. This is a longer text which

5276

# describes the expression, e.g. when hovered over it in a UI.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5277

},

5278

"members": [ # Specifies the identities requesting access for a Cloud Platform resource.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5279

# `members` can have the following values:

5280

#

5281

# * `allUsers`: A special identifier that represents anyone who is

5282

# on the internet; with or without a Google account.

5283

#

5284

# * `allAuthenticatedUsers`: A special identifier that represents anyone

5285

# who is authenticated with a Google account or a service account.

5286

#

5287

# * `user:{emailid}`: An email address that represents a specific Google

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5288

# account. For example, `alice@example.com` .

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5289

#

5290

#

5291

# * `serviceAccount:{emailid}`: An email address that represents a service

5292

# account. For example, `my-other-app@appspot.gserviceaccount.com`.

5293

#

5294

# * `group:{emailid}`: An email address that represents a Google group.

5295

# For example, `admins@example.com`.

5296

#

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5297

# * `deleted:user:{emailid}?uid={uniqueid}`: An email address (plus unique

5298

# identifier) representing a user that has been recently deleted. For

5299

# example, `alice@example.com?uid=123456789012345678901`. If the user is

5300

# recovered, this value reverts to `user:{emailid}` and the recovered user

5301

# retains the role in the binding.

5302

#

5303

# * `deleted:serviceAccount:{emailid}?uid={uniqueid}`: An email address (plus

5304

# unique identifier) representing a service account that has been recently

5305

# deleted. For example,

5306

# `my-other-app@appspot.gserviceaccount.com?uid=123456789012345678901`.

5307

# If the service account is undeleted, this value reverts to

5308

# `serviceAccount:{emailid}` and the undeleted service account retains the

5309

# role in the binding.

5310

#

5311

# * `deleted:group:{emailid}?uid={uniqueid}`: An email address (plus unique

5312

# identifier) representing a Google group that has been recently

5313

# deleted. For example, `admins@example.com?uid=123456789012345678901`. If

5314

# the group is recovered, this value reverts to `group:{emailid}` and the

5315

# recovered group retains the role in the binding.

5316

#

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5317

#

5318

# * `domain:{domain}`: The G Suite domain (primary) that represents all the

5319

# users of that domain. For example, `google.com` or `example.com`.

5320

#

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5321

"A String",

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5322

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5323

},

5324

],

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

}</pre>

</div>

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5329

<code class="details" id="testIamPermissions">testIamPermissions(resource, body=None, x__xgafv=None)</code>

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5330

<pre>Returns permissions that a caller has on the specified resource.

5331

If the resource does not exist, this will return an empty set of

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5332

permissions, not a `NOT_FOUND` error.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5333

5334

Note: This operation is designed to be used for building permission-aware

5335

UIs and command-line tools, not for authorization checking. This operation

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5336

may "fail open" without warning.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5337

5338

Args:

5339

resource: string, REQUIRED: The resource for which the policy detail is being requested.

5340

See the operation documentation for the appropriate value for this field. (required)

Dan O'Meara

2020-05-01 07:42:23 -0700

[diff] [blame]

5341

body: object, The request body.

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5342

The object takes the form of:

5343

5344

{ # Request message for `TestIamPermissions` method.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5345

"permissions": [ # The set of permissions to check for the `resource`. Permissions with

5346

# wildcards (such as '*' or 'storage.*') are not allowed. For more

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5347

# information see

5348

# [IAM Overview](https://cloud.google.com/iam/docs/overview#permissions).

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5349

"A String",

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

],

}

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

5360

5361

{ # Response message for `TestIamPermissions` method.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5362

"permissions": [ # A subset of `TestPermissionsRequest.permissions` that the caller is

Bu Sun Kim

2019-06-14 16:50:42 -0700

[diff] [blame]

5363

# allowed.

Bu Sun Kim

2020-05-20 12:08:20 -0700

[diff] [blame]

5364

"A String",

Bu Sun Kim