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

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

[diff] [blame]

76

<h2>Instance Methods</h2>

77

Dan O'Meara

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

[diff] [blame]

78

<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]

79

<p class="firstline">Creates a new version of a model from a trained TensorFlow model.</p>

80

Thomas Coffee

2f24537

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

[diff] [blame]

81

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

Sai Cheemalapati

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

[diff] [blame]

82

<p class="firstline">Deletes a model version.</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">Gets information about a model version.</p>

86

Bu Sun Kim

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

[diff] [blame]

87

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

Sai Cheemalapati

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

[diff] [blame]

88

<p class="firstline">Gets basic information about all the versions of a model.</p>

89

90

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

91

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

92

Dan O'Meara

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

[diff] [blame]

93

<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]

94

<p class="firstline">Updates the specified Version resource.</p>

95

96

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

Sai Cheemalapati

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

[diff] [blame]

97

<p class="firstline">Designates a version to be the default for the model.</p>

98

<h3>Method Details</h3>

99

Dan O'Meara

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

[diff] [blame]

100

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

Sai Cheemalapati

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

[diff] [blame]

101

<pre>Creates a new version of a model from a trained TensorFlow model.

102

103

If the version created in the cloud by this call is the first deployed

104

version of the specified model, it will be made the default version of the

105

model. When you add a version to a model that already has one or more

106

versions, the default version does not automatically change. If you want a

107

new version to be the default, you must call

Dan O'Meara

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

[diff] [blame]

108

projects.models.versions.setDefault.

Sai Cheemalapati

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

[diff] [blame]

109

110

Args:

Bu Sun Kim

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

[diff] [blame]

111

parent: string, Required. The name of the model. (required)

Dan O'Meara

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

[diff] [blame]

112

body: object, The request body.

Sai Cheemalapati

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

[diff] [blame]

113

The object takes the form of:

114

115

{ # Represents a version of the model.

116

#

117

# Each version is a trained model deployed in the cloud, ready to handle

118

# prediction requests. A model can have multiple versions. You can get

119

# information about all of the versions of a given model by calling

Dan O'Meara

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

[diff] [blame]

120

# projects.models.versions.list.

Bu Sun Kim

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

[diff] [blame]

121

"acceleratorConfig": { # Represents a hardware accelerator request config. # Optional. Accelerator config for using GPUs for online prediction (beta).

122

# Only specify this field if you have specified a Compute Engine (N1) machine

123

# type in the `machineType` field. Learn more about [using GPUs for online

124

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

125

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

126

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

127

# [accelerators for online

128

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

129

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

130

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

131

},

132

"isDefault": True or False, # Output only. If true, this version will be used to handle prediction

133

# requests that do not specify a version.

134

#

135

# You can change the default version by calling

136

# projects.methods.versions.setDefault.

Bu Sun Kim

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

[diff] [blame]

137

"manualScaling": { # Options for manually scaling a model. # Manually select the number of nodes to use for serving the

138

# model. You should generally use `auto_scaling` with an appropriate

139

# `min_nodes` instead, but this option is available if you want more

140

# predictable billing. Beware that latency and error rates will increase

141

# if the traffic exceeds that capability of the system to serve it based

142

# on the selected number of nodes.

143

"nodes": 42, # The number of nodes to allocate for this model. These nodes are always up,

144

# starting from the time the model is deployed, so the cost of operating

145

# this model will be proportional to `nodes` * number of hours since

146

# last billing cycle plus the cost for each prediction performed.

Dan O'Meara

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

[diff] [blame]

147

},

Bu Sun Kim

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

[diff] [blame]

148

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

Bu Sun Kim

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

[diff] [blame]

149

"name": "A String", # Required. The name specified for the version when it was created.

150

#

151

# The version name must be unique within the model it is created in.

152

"serviceAccount": "A String", # Optional. Specifies the service account for resource access control.

153

"pythonVersion": "A String", # Required. The version of Python used in prediction.

154

#

155

# The following Python versions are available:

156

#

157

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

158

# later.

159

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

160

# from '1.4' to '1.14'.

161

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

162

# earlier.

163

#

164

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

165

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

166

"lastUseTime": "A String", # Output only. The time the version was last used for prediction.

167

"predictionClass": "A String", # Optional. The fully qualified name

Dan O'Meara

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

[diff] [blame]

168

# (<var>module_name</var>.<var>class_name</var>) of a class that implements

Bu Sun Kim

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

[diff] [blame]

169

# the Predictor interface described in this reference field. The module

170

# containing this class should be included in a package provided to the

171

# [`packageUris` field](#Version.FIELDS.package_uris).

Sai Cheemalapati

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

[diff] [blame]

172

#

Bu Sun Kim

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

[diff] [blame]

173

# Specify this field if and only if you are deploying a [custom prediction

174

# routine (beta)](/ml-engine/docs/tensorflow/custom-prediction-routines).

175

# If you specify this field, you must set

Dan O'Meara

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

[diff] [blame]

176

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater and

177

# you must set `machineType` to a [legacy (MLS1)

178

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

Bu Sun Kim

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

[diff] [blame]

179

#

180

# The following code sample provides the Predictor interface:

181

#

Bu Sun Kim

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

[diff] [blame]

182

# <pre style="max-width: 626px;">

Bu Sun Kim

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

[diff] [blame]

183

# class Predictor(object):

Bu Sun Kim

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

[diff] [blame]

184

# """Interface for constructing custom predictors."""

Bu Sun Kim

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

[diff] [blame]

185

#

186

# def predict(self, instances, **kwargs):

Bu Sun Kim

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

[diff] [blame]

187

# """Performs custom prediction.

Bu Sun Kim

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

[diff] [blame]

188

#

189

# Instances are the decoded values from the request. They have already

190

# been deserialized from JSON.

191

#

192

# Args:

193

# instances: A list of prediction input instances.

194

# **kwargs: A dictionary of keyword args provided as additional

195

# fields on the predict request body.

196

#

197

# Returns:

198

# A list of outputs containing the prediction results. This list must

199

# be JSON serializable.

Bu Sun Kim

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

[diff] [blame]

200

# """

Bu Sun Kim

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

[diff] [blame]

201

# raise NotImplementedError()

202

#

203

# @classmethod

204

# def from_path(cls, model_dir):

Bu Sun Kim

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

[diff] [blame]

205

# """Creates an instance of Predictor using the given path.

Bu Sun Kim

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

[diff] [blame]

206

#

207

# Loading of the predictor should be done in this method.

208

#

209

# Args:

210

# model_dir: The local directory that contains the exported model

211

# file along with any additional files uploaded when creating the

# version resource.

#

# Returns:

# An instance implementing this Predictor class.

Bu Sun Kim

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

[diff] [blame]

216

# """

Bu Sun Kim

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

[diff] [blame]

217

# raise NotImplementedError()

Dan O'Meara

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

[diff] [blame]

218

# </pre>

Bu Sun Kim

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

[diff] [blame]

219

#

220

# Learn more about [the Predictor interface and custom prediction

221

# routines](/ml-engine/docs/tensorflow/custom-prediction-routines).

Bu Sun Kim

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

[diff] [blame]

222

"deploymentUri": "A String", # Required. The Cloud Storage location of the trained model used to

223

# create the version. See the

224

# [guide to model

225

# deployment](/ml-engine/docs/tensorflow/deploying-models) for more

226

# information.

227

#

228

# When passing Version to

229

# projects.models.versions.create

230

# the model service uses the specified location as the source of the model.

231

# Once deployed, the model version is hosted by the prediction service, so

232

# this location is useful only as a historical record.

233

# The total number of model files can't exceed 1000.

Bu Sun Kim

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

[diff] [blame]

234

"packageUris": [ # Optional. Cloud Storage paths (`gs://…`) of packages for [custom

Bu Sun Kim

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

[diff] [blame]

235

# prediction routines](/ml-engine/docs/tensorflow/custom-prediction-routines)

236

# or [scikit-learn pipelines with custom

237

# code](/ml-engine/docs/scikit/exporting-for-prediction#custom-pipeline-code).

238

#

239

# For a custom prediction routine, one of these packages must contain your

240

# Predictor class (see

241

# [`predictionClass`](#Version.FIELDS.prediction_class)). Additionally,

242

# include any dependencies used by your Predictor or scikit-learn pipeline

243

# uses that are not already included in your selected [runtime

244

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

245

#

246

# If you specify this field, you must also set

247

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater.

Bu Sun Kim

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

[diff] [blame]

248

"A String",

Bu Sun Kim

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

[diff] [blame]

249

],

Bu Sun Kim

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

[diff] [blame]

250

"explanationConfig": { # Message holding configuration options for explaining model predictions. # Optional. Configures explainability features on the model's version.

251

# Some explanation features require additional metadata to be loaded

252

# as part of the model payload.

253

# There are two feature attribution methods supported for TensorFlow models:

254

# integrated gradients and sampled Shapley.

255

# [Learn more about feature

256

# attributions.](/ai-platform/prediction/docs/ai-explanations/overview)

257

"integratedGradientsAttribution": { # Attributes credit by computing the Aumann-Shapley value taking advantage # Attributes credit by computing the Aumann-Shapley value taking advantage

258

# of the model's fully differentiable structure. Refer to this paper for

259

# more details: http://proceedings.mlr.press/v70/sundararajan17a.html

260

# of the model's fully differentiable structure. Refer to this paper for

261

# more details: https://arxiv.org/abs/1703.01365

262

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

263

# A good value to start is 50 and gradually increase until the

264

# sum to diff property is met within the desired error range.

265

},

266

"sampledShapleyAttribution": { # An attribution method that approximates Shapley values for features that # An attribution method that approximates Shapley values for features that

267

# contribute to the label being predicted. A sampling strategy is used to

268

# approximate the value rather than considering all subsets of features.

269

# contribute to the label being predicted. A sampling strategy is used to

270

# approximate the value rather than considering all subsets of features.

271

"numPaths": 42, # The number of feature permutations to consider when approximating the

272

# Shapley values.

273

},

274

"xraiAttribution": { # Attributes credit by computing the XRAI taking advantage # Attributes credit by computing the XRAI taking advantage

275

# of the model's fully differentiable structure. Refer to this paper for

276

# more details: https://arxiv.org/abs/1906.02825

277

# Currently only implemented for models with natural image inputs.

278

# of the model's fully differentiable structure. Refer to this paper for

279

# more details: https://arxiv.org/abs/1906.02825

280

# Currently only implemented for models with natural image inputs.

281

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

282

# A good value to start is 50 and gradually increase until the

283

# sum to diff property is met within the desired error range.

284

},

285

},

Bu Sun Kim

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

[diff] [blame]

286

"autoScaling": { # Options for automatically scaling a model. # Automatically scale the number of nodes used to serve the model in

Dan O'Meara

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

[diff] [blame]

287

# response to increases and decreases in traffic. Care should be

Bu Sun Kim

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

[diff] [blame]

288

# taken to ramp up traffic according to the model's ability to scale

Dan O'Meara

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

[diff] [blame]

289

# or you will start seeing increases in latency and 429 response codes.

290

#

291

# Note that you cannot use AutoScaling if your version uses

292

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use specify

293

# `manual_scaling`.

Bu Sun Kim

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

[diff] [blame]

294

"minNodes": 42, # Optional. The minimum number of nodes to allocate for this model. These

Dan O'Meara

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

[diff] [blame]

295

# nodes are always up, starting from the time the model is deployed.

296

# Therefore, the cost of operating this model will be at least

297

# `rate` * `min_nodes` * number of hours since last billing cycle,

298

# where `rate` is the cost per node-hour as documented in the

299

# [pricing guide](/ml-engine/docs/pricing),

300

# even if no predictions are performed. There is additional cost for each

301

# prediction performed.

302

#

303

# Unlike manual scaling, if the load gets too heavy for the nodes

304

# that are up, the service will automatically add nodes to handle the

305

# increased load as well as scale back as traffic drops, always maintaining

306

# at least `min_nodes`. You will be charged for the time in which additional

307

# nodes are used.

308

#

309

# If `min_nodes` is not specified and AutoScaling is used with a [legacy

310

# (MLS1) machine type](/ml-engine/docs/machine-types-online-prediction),

311

# `min_nodes` defaults to 0, in which case, when traffic to a model stops

312

# (and after a cool-down period), nodes will be shut down and no charges will

313

# be incurred until traffic to the model resumes.

314

#

315

# If `min_nodes` is not specified and AutoScaling is used with a [Compute

316

# Engine (N1) machine type](/ml-engine/docs/machine-types-online-prediction),

317

# `min_nodes` defaults to 1. `min_nodes` must be at least 1 for use with a

318

# Compute Engine machine type.

319

#

320

# Note that you cannot use AutoScaling if your version uses

321

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use

322

# ManualScaling.

323

#

324

# You can set `min_nodes` when creating the model version, and you can also

325

# update `min_nodes` for an existing version:

326

# <pre>

327

# update_body.json:

328

# {

Bu Sun Kim

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

[diff] [blame]

329

# 'autoScaling': {

330

# 'minNodes': 5

Dan O'Meara

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

[diff] [blame]

# }

# }

# </pre>

# HTTP request:

Bu Sun Kim

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

[diff] [blame]

335

# <pre style="max-width: 626px;">

Dan O'Meara

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

[diff] [blame]

336

# PATCH

337

# https://ml.googleapis.com/v1/{name=projects/*/models/*/versions/*}?update_mask=autoScaling.minNodes

338

# -d @./update_body.json

339

# </pre>

340

},

Bu Sun Kim

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

[diff] [blame]

341

"createTime": "A String", # Output only. The time the version was created.

Bu Sun Kim

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

[diff] [blame]

342

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

343

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

344

# are arbitrary strings that you supply.

345

# For more information, see the documentation on

346

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

347

"a_key": "A String",

348

},

Bu Sun Kim

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

[diff] [blame]

349

"requestLoggingConfig": { # Configuration for logging request-response pairs to a BigQuery table. # Optional. *Only* specify this field in a

Dan O'Meara

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

[diff] [blame]

350

# projects.models.versions.patch

351

# request. Specifying it in a

352

# projects.models.versions.create

353

# request has no effect.

354

#

355

# Configures the request-response pair logging on predictions from this

356

# Version.

357

# Online prediction requests to a model version and the responses to these

358

# requests are converted to raw strings and saved to the specified BigQuery

359

# table. Logging is constrained by [BigQuery quotas and

360

# limits](/bigquery/quotas). If your project exceeds BigQuery quotas or limits,

361

# AI Platform Prediction does not log request-response pairs, but it continues

362

# to serve predictions.

363

#

364

# If you are using [continuous

365

# evaluation](/ml-engine/docs/continuous-evaluation/), you do not need to

366

# specify this configuration manually. Setting up continuous evaluation

367

# automatically enables logging of request-response pairs.

Bu Sun Kim

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

[diff] [blame]

368

"bigqueryTableName": "A String", # Required. Fully qualified BigQuery table name in the following format:

369

# "<var>project_id</var>.<var>dataset_name</var>.<var>table_name</var>"

Dan O'Meara

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

[diff] [blame]

370

#

Bu Sun Kim

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

[diff] [blame]

371

# The specified table must already exist, and the "Cloud ML Service Agent"

Dan O'Meara

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

[diff] [blame]

372

# for your project must have permission to write to it. The table must have

373

# the following [schema](/bigquery/docs/schemas):

374

#

375

# <table>

Bu Sun Kim

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

[diff] [blame]

376

# <tr><th>Field name</th><th style="display: table-cell">Type</th>

377

# <th style="display: table-cell">Mode</th></tr>

Dan O'Meara

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

[diff] [blame]

378

# <tr><td>model</td><td>STRING</td><td>REQUIRED</td></tr>

379

# <tr><td>model_version</td><td>STRING</td><td>REQUIRED</td></tr>

380

# <tr><td>time</td><td>TIMESTAMP</td><td>REQUIRED</td></tr>

381

# <tr><td>raw_data</td><td>STRING</td><td>REQUIRED</td></tr>

382

# <tr><td>raw_prediction</td><td>STRING</td><td>NULLABLE</td></tr>

383

# <tr><td>groundtruth</td><td>STRING</td><td>NULLABLE</td></tr>

384

# </table>

Bu Sun Kim

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

[diff] [blame]

385

"samplingPercentage": 3.14, # Percentage of requests to be logged, expressed as a fraction from 0 to 1.

386

# For example, if you want to log 10% of requests, enter `0.1`. The sampling

387

# window is the lifetime of the model version. Defaults to 0.

388

},

389

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

390

"machineType": "A String", # Optional. The type of machine on which to serve the model. Currently only

391

# applies to online prediction service. If this field is not specified, it

392

# defaults to `mls1-c1-m2`.

393

#

394

# Online prediction supports the following machine types:

#

# * `mls1-c1-m2`

# * `mls1-c4-m2`

# * `n1-standard-2`

# * `n1-standard-4`

# * `n1-standard-8`

# * `n1-standard-16`

# * `n1-standard-32`

# * `n1-highmem-2`

# * `n1-highmem-4`

# * `n1-highmem-8`

# * `n1-highmem-16`

# * `n1-highmem-32`

# * `n1-highcpu-2`

# * `n1-highcpu-4`

# * `n1-highcpu-8`

# * `n1-highcpu-16`

# * `n1-highcpu-32`

#

# `mls1-c1-m2` is generally available. All other machine types are available

415

# in beta. Learn more about the [differences between machine

416

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

417

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

418

#

419

# For more information, see the

420

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

421

# [how to manage runtime versions](/ml-engine/docs/versioning).

422

"description": "A String", # Optional. The description specified for the version when it was created.

423

"framework": "A String", # Optional. The machine learning framework AI Platform uses to train

424

# this version of the model. Valid values are `TENSORFLOW`, `SCIKIT_LEARN`,

425

# `XGBOOST`. If you do not specify a framework, AI Platform

426

# will analyze files in the deployment_uri to determine a framework. If you

427

# choose `SCIKIT_LEARN` or `XGBOOST`, you must also set the runtime version

428

# of the model to 1.4 or greater.

429

#

430

# Do **not** specify a framework if you're deploying a [custom

431

# prediction routine](/ml-engine/docs/tensorflow/custom-prediction-routines).

432

#

433

# If you specify a [Compute Engine (N1) machine

434

# type](/ml-engine/docs/machine-types-online-prediction) in the

435

# `machineType` field, you must specify `TENSORFLOW`

436

# for the framework.

437

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

438

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

439

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

440

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

441

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

442

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

443

# ensure that their change will be applied to the model as intended.

Sai Cheemalapati

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

[diff] [blame]

444

}

445

446

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

453

454

{ # This resource represents a long-running operation that is the result of a

455

# network API call.

Bu Sun Kim

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

[diff] [blame]

456

"name": "A String", # The server-assigned name, which is only unique within the same service that

Sai Cheemalapati

4ba8c23

2017-06-06 18:46:08 -0400

[diff] [blame]

457

# originally returns it. If you use the default HTTP mapping, the

Bu Sun Kim

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

[diff] [blame]

458

# `name` should be a resource name ending with `operations/{unique_id}`.

Bu Sun Kim

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

[diff] [blame]

459

"error": { # The `Status` type defines a logical error model that is suitable for # The error result of the operation in case of failure or cancellation.

460

# different programming environments, including REST APIs and RPC APIs. It is

461

# used by [gRPC](https://github.com/grpc). Each `Status` message contains

462

# three pieces of data: error code, error message, and error details.

463

#

464

# You can find out more about this error model and how to work with it in the

465

# [API Design Guide](https://cloud.google.com/apis/design/errors).

Bu Sun Kim

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

[diff] [blame]

466

"code": 42, # The status code, which should be an enum value of google.rpc.Code.

467

"message": "A String", # A developer-facing error message, which should be in English. Any

468

# user-facing error message should be localized and sent in the

469

# google.rpc.Status.details field, or localized by the client.

Bu Sun Kim

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

[diff] [blame]

470

"details": [ # A list of messages that carry the error details. There is a common set of

471

# message types for APIs to use.

472

{

473

"a_key": "", # Properties of the object. Contains field @type with type URL.

474

},

475

],

Bu Sun Kim

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

[diff] [blame]

476

},

477

"metadata": { # Service-specific metadata associated with the operation. It typically

478

# contains progress information and common metadata such as create time.

479

# Some services might not provide such metadata. Any method that returns a

480

# long-running operation should document the metadata type, if any.

481

"a_key": "", # Properties of the object. Contains field @type with type URL.

482

},

483

"done": True or False, # If the value is `false`, it means the operation is still in progress.

484

# If `true`, the operation is completed, and either `error` or `response` is

485

# available.

Bu Sun Kim

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

[diff] [blame]

486

"response": { # The normal response of the operation in case of success. If the original

487

# method returns no data on success, such as `Delete`, the response is

488

# `google.protobuf.Empty`. If the original method is standard

489

# `Get`/`Create`/`Update`, the response should be the resource. For other

490

# methods, the response should have the type `XxxResponse`, where `Xxx`

491

# is the original method name. For example, if the original method name

492

# is `TakeSnapshot()`, the inferred response type is

493

# `TakeSnapshotResponse`.

494

"a_key": "", # Properties of the object. Contains field @type with type URL.

495

},

Sai Cheemalapati

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

[diff] [blame]

}</pre>

</div>

Thomas Coffee

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

[diff] [blame]

500

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

Sai Cheemalapati

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

[diff] [blame]

501

<pre>Deletes a model version.

502

503

Each model can have multiple versions deployed and in use at any given

504

time. Use this method to remove a single version.

505

506

Note: You cannot delete the version that is set as the default version

507

of the model unless it is the only remaining version.

508

509

Args:

510

name: string, Required. The name of the version. You can get the names of all the

511

versions of a model by calling

Dan O'Meara

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

[diff] [blame]

512

projects.models.versions.list. (required)

Sai Cheemalapati

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

[diff] [blame]

513

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

520

521

{ # This resource represents a long-running operation that is the result of a

522

# network API call.

Bu Sun Kim

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

[diff] [blame]

523

"name": "A String", # The server-assigned name, which is only unique within the same service that

Sai Cheemalapati

4ba8c23

2017-06-06 18:46:08 -0400

[diff] [blame]

524

# originally returns it. If you use the default HTTP mapping, the

Bu Sun Kim

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

[diff] [blame]

525

# `name` should be a resource name ending with `operations/{unique_id}`.

Bu Sun Kim

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

[diff] [blame]

526

"error": { # The `Status` type defines a logical error model that is suitable for # The error result of the operation in case of failure or cancellation.

527

# different programming environments, including REST APIs and RPC APIs. It is

528

# used by [gRPC](https://github.com/grpc). Each `Status` message contains

529

# three pieces of data: error code, error message, and error details.

530

#

531

# You can find out more about this error model and how to work with it in the

532

# [API Design Guide](https://cloud.google.com/apis/design/errors).

Bu Sun Kim

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

[diff] [blame]

533

"code": 42, # The status code, which should be an enum value of google.rpc.Code.

534

"message": "A String", # A developer-facing error message, which should be in English. Any

535

# user-facing error message should be localized and sent in the

536

# google.rpc.Status.details field, or localized by the client.

Bu Sun Kim

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

[diff] [blame]

537

"details": [ # A list of messages that carry the error details. There is a common set of

538

# message types for APIs to use.

539

{

540

"a_key": "", # Properties of the object. Contains field @type with type URL.

541

},

542

],

Bu Sun Kim

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

[diff] [blame]

543

},

544

"metadata": { # Service-specific metadata associated with the operation. It typically

545

# contains progress information and common metadata such as create time.

546

# Some services might not provide such metadata. Any method that returns a

547

# long-running operation should document the metadata type, if any.

548

"a_key": "", # Properties of the object. Contains field @type with type URL.

549

},

550

"done": True or False, # If the value is `false`, it means the operation is still in progress.

551

# If `true`, the operation is completed, and either `error` or `response` is

552

# available.

Bu Sun Kim

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

[diff] [blame]

553

"response": { # The normal response of the operation in case of success. If the original

554

# method returns no data on success, such as `Delete`, the response is

555

# `google.protobuf.Empty`. If the original method is standard

556

# `Get`/`Create`/`Update`, the response should be the resource. For other

557

# methods, the response should have the type `XxxResponse`, where `Xxx`

558

# is the original method name. For example, if the original method name

559

# is `TakeSnapshot()`, the inferred response type is

560

# `TakeSnapshotResponse`.

561

"a_key": "", # Properties of the object. Contains field @type with type URL.

562

},

Sai Cheemalapati

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

[diff] [blame]

}</pre>

</div>

Thomas Coffee

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

[diff] [blame]

567

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

Sai Cheemalapati

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

[diff] [blame]

568

<pre>Gets information about a model version.

569

570

Models can have multiple versions. You can call

Dan O'Meara

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

[diff] [blame]

571

projects.models.versions.list

Sai Cheemalapati

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

[diff] [blame]

572

to get the same information that this method returns for all of the

573

versions of a model.

574

575

Args:

Bu Sun Kim

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

[diff] [blame]

576

name: string, Required. The name of the version. (required)

Sai Cheemalapati

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

[diff] [blame]

577

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

584

585

{ # Represents a version of the model.

586

#

587

# Each version is a trained model deployed in the cloud, ready to handle

588

# prediction requests. A model can have multiple versions. You can get

589

# information about all of the versions of a given model by calling

Dan O'Meara

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

[diff] [blame]

590

# projects.models.versions.list.

Bu Sun Kim

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

[diff] [blame]

591

"acceleratorConfig": { # Represents a hardware accelerator request config. # Optional. Accelerator config for using GPUs for online prediction (beta).

592

# Only specify this field if you have specified a Compute Engine (N1) machine

593

# type in the `machineType` field. Learn more about [using GPUs for online

594

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

595

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

596

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

597

# [accelerators for online

598

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

599

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

600

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

601

},

602

"isDefault": True or False, # Output only. If true, this version will be used to handle prediction

603

# requests that do not specify a version.

604

#

605

# You can change the default version by calling

606

# projects.methods.versions.setDefault.

Bu Sun Kim

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

[diff] [blame]

607

"manualScaling": { # Options for manually scaling a model. # Manually select the number of nodes to use for serving the

608

# model. You should generally use `auto_scaling` with an appropriate

609

# `min_nodes` instead, but this option is available if you want more

610

# predictable billing. Beware that latency and error rates will increase

611

# if the traffic exceeds that capability of the system to serve it based

612

# on the selected number of nodes.

613

"nodes": 42, # The number of nodes to allocate for this model. These nodes are always up,

614

# starting from the time the model is deployed, so the cost of operating

615

# this model will be proportional to `nodes` * number of hours since

616

# last billing cycle plus the cost for each prediction performed.

Dan O'Meara

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

[diff] [blame]

617

},

Bu Sun Kim

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

[diff] [blame]

618

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

Bu Sun Kim

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

[diff] [blame]

619

"name": "A String", # Required. The name specified for the version when it was created.

620

#

621

# The version name must be unique within the model it is created in.

622

"serviceAccount": "A String", # Optional. Specifies the service account for resource access control.

623

"pythonVersion": "A String", # Required. The version of Python used in prediction.

624

#

625

# The following Python versions are available:

626

#

627

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

628

# later.

629

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

630

# from '1.4' to '1.14'.

631

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

632

# earlier.

633

#

634

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

635

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

636

"lastUseTime": "A String", # Output only. The time the version was last used for prediction.

637

"predictionClass": "A String", # Optional. The fully qualified name

Dan O'Meara

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

[diff] [blame]

638

# (<var>module_name</var>.<var>class_name</var>) of a class that implements

Bu Sun Kim

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

[diff] [blame]

639

# the Predictor interface described in this reference field. The module

640

# containing this class should be included in a package provided to the

641

# [`packageUris` field](#Version.FIELDS.package_uris).

Sai Cheemalapati

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

[diff] [blame]

642

#

Bu Sun Kim

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

[diff] [blame]

643

# Specify this field if and only if you are deploying a [custom prediction

644

# routine (beta)](/ml-engine/docs/tensorflow/custom-prediction-routines).

645

# If you specify this field, you must set

Dan O'Meara

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

[diff] [blame]

646

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater and

647

# you must set `machineType` to a [legacy (MLS1)

648

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

Bu Sun Kim

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

[diff] [blame]

649

#

650

# The following code sample provides the Predictor interface:

651

#

Bu Sun Kim

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

[diff] [blame]

652

# <pre style="max-width: 626px;">

Bu Sun Kim

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

[diff] [blame]

653

# class Predictor(object):

Bu Sun Kim

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

[diff] [blame]

654

# """Interface for constructing custom predictors."""

Bu Sun Kim

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

[diff] [blame]

655

#

656

# def predict(self, instances, **kwargs):

Bu Sun Kim

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

[diff] [blame]

657

# """Performs custom prediction.

Bu Sun Kim

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

[diff] [blame]

658

#

659

# Instances are the decoded values from the request. They have already

660

# been deserialized from JSON.

661

#

662

# Args:

663

# instances: A list of prediction input instances.

664

# **kwargs: A dictionary of keyword args provided as additional

665

# fields on the predict request body.

666

#

667

# Returns:

668

# A list of outputs containing the prediction results. This list must

669

# be JSON serializable.

Bu Sun Kim

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

[diff] [blame]

670

# """

Bu Sun Kim

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

[diff] [blame]

671

# raise NotImplementedError()

672

#

673

# @classmethod

674

# def from_path(cls, model_dir):

Bu Sun Kim

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

[diff] [blame]

675

# """Creates an instance of Predictor using the given path.

Bu Sun Kim

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

[diff] [blame]

676

#

677

# Loading of the predictor should be done in this method.

678

#

679

# Args:

680

# model_dir: The local directory that contains the exported model

681

# file along with any additional files uploaded when creating the

# version resource.

#

# Returns:

# An instance implementing this Predictor class.

Bu Sun Kim

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

[diff] [blame]

686

# """

Bu Sun Kim

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

[diff] [blame]

687

# raise NotImplementedError()

Dan O'Meara

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

[diff] [blame]

688

# </pre>

Bu Sun Kim

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

[diff] [blame]

689

#

690

# Learn more about [the Predictor interface and custom prediction

691

# routines](/ml-engine/docs/tensorflow/custom-prediction-routines).

Bu Sun Kim

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

[diff] [blame]

692

"deploymentUri": "A String", # Required. The Cloud Storage location of the trained model used to

693

# create the version. See the

694

# [guide to model

695

# deployment](/ml-engine/docs/tensorflow/deploying-models) for more

696

# information.

697

#

698

# When passing Version to

699

# projects.models.versions.create

700

# the model service uses the specified location as the source of the model.

701

# Once deployed, the model version is hosted by the prediction service, so

702

# this location is useful only as a historical record.

703

# The total number of model files can't exceed 1000.

Bu Sun Kim

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

[diff] [blame]

704

"packageUris": [ # Optional. Cloud Storage paths (`gs://…`) of packages for [custom

Bu Sun Kim

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

[diff] [blame]

705

# prediction routines](/ml-engine/docs/tensorflow/custom-prediction-routines)

706

# or [scikit-learn pipelines with custom

707

# code](/ml-engine/docs/scikit/exporting-for-prediction#custom-pipeline-code).

708

#

709

# For a custom prediction routine, one of these packages must contain your

710

# Predictor class (see

711

# [`predictionClass`](#Version.FIELDS.prediction_class)). Additionally,

712

# include any dependencies used by your Predictor or scikit-learn pipeline

713

# uses that are not already included in your selected [runtime

714

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

715

#

716

# If you specify this field, you must also set

717

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater.

Bu Sun Kim

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

[diff] [blame]

718

"A String",

Bu Sun Kim

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

[diff] [blame]

719

],

Bu Sun Kim

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

[diff] [blame]

720

"explanationConfig": { # Message holding configuration options for explaining model predictions. # Optional. Configures explainability features on the model's version.

721

# Some explanation features require additional metadata to be loaded

722

# as part of the model payload.

723

# There are two feature attribution methods supported for TensorFlow models:

724

# integrated gradients and sampled Shapley.

725

# [Learn more about feature

726

# attributions.](/ai-platform/prediction/docs/ai-explanations/overview)

727

"integratedGradientsAttribution": { # Attributes credit by computing the Aumann-Shapley value taking advantage # Attributes credit by computing the Aumann-Shapley value taking advantage

728

# of the model's fully differentiable structure. Refer to this paper for

729

# more details: http://proceedings.mlr.press/v70/sundararajan17a.html

730

# of the model's fully differentiable structure. Refer to this paper for

731

# more details: https://arxiv.org/abs/1703.01365

732

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

733

# A good value to start is 50 and gradually increase until the

734

# sum to diff property is met within the desired error range.

735

},

736

"sampledShapleyAttribution": { # An attribution method that approximates Shapley values for features that # An attribution method that approximates Shapley values for features that

737

# contribute to the label being predicted. A sampling strategy is used to

738

# approximate the value rather than considering all subsets of features.

739

# contribute to the label being predicted. A sampling strategy is used to

740

# approximate the value rather than considering all subsets of features.

741

"numPaths": 42, # The number of feature permutations to consider when approximating the

742

# Shapley values.

743

},

744

"xraiAttribution": { # Attributes credit by computing the XRAI taking advantage # Attributes credit by computing the XRAI taking advantage

745

# of the model's fully differentiable structure. Refer to this paper for

746

# more details: https://arxiv.org/abs/1906.02825

747

# Currently only implemented for models with natural image inputs.

748

# of the model's fully differentiable structure. Refer to this paper for

749

# more details: https://arxiv.org/abs/1906.02825

750

# Currently only implemented for models with natural image inputs.

751

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

752

# A good value to start is 50 and gradually increase until the

753

# sum to diff property is met within the desired error range.

754

},

755

},

Bu Sun Kim

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

[diff] [blame]

756

"autoScaling": { # Options for automatically scaling a model. # Automatically scale the number of nodes used to serve the model in

Dan O'Meara

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

[diff] [blame]

757

# response to increases and decreases in traffic. Care should be

Bu Sun Kim

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

[diff] [blame]

758

# taken to ramp up traffic according to the model's ability to scale

Dan O'Meara

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

[diff] [blame]

759

# or you will start seeing increases in latency and 429 response codes.

760

#

761

# Note that you cannot use AutoScaling if your version uses

762

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use specify

763

# `manual_scaling`.

Bu Sun Kim

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

[diff] [blame]

764

"minNodes": 42, # Optional. The minimum number of nodes to allocate for this model. These

Dan O'Meara

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

[diff] [blame]

765

# nodes are always up, starting from the time the model is deployed.

766

# Therefore, the cost of operating this model will be at least

767

# `rate` * `min_nodes` * number of hours since last billing cycle,

768

# where `rate` is the cost per node-hour as documented in the

769

# [pricing guide](/ml-engine/docs/pricing),

770

# even if no predictions are performed. There is additional cost for each

771

# prediction performed.

772

#

773

# Unlike manual scaling, if the load gets too heavy for the nodes

774

# that are up, the service will automatically add nodes to handle the

775

# increased load as well as scale back as traffic drops, always maintaining

776

# at least `min_nodes`. You will be charged for the time in which additional

777

# nodes are used.

778

#

779

# If `min_nodes` is not specified and AutoScaling is used with a [legacy

780

# (MLS1) machine type](/ml-engine/docs/machine-types-online-prediction),

781

# `min_nodes` defaults to 0, in which case, when traffic to a model stops

782

# (and after a cool-down period), nodes will be shut down and no charges will

783

# be incurred until traffic to the model resumes.

784

#

785

# If `min_nodes` is not specified and AutoScaling is used with a [Compute

786

# Engine (N1) machine type](/ml-engine/docs/machine-types-online-prediction),

787

# `min_nodes` defaults to 1. `min_nodes` must be at least 1 for use with a

788

# Compute Engine machine type.

789

#

790

# Note that you cannot use AutoScaling if your version uses

791

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use

792

# ManualScaling.

793

#

794

# You can set `min_nodes` when creating the model version, and you can also

795

# update `min_nodes` for an existing version:

796

# <pre>

797

# update_body.json:

798

# {

Bu Sun Kim

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

[diff] [blame]

799

# 'autoScaling': {

800

# 'minNodes': 5

Dan O'Meara

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

[diff] [blame]

# }

# }

# </pre>

# HTTP request:

Bu Sun Kim

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

[diff] [blame]

805

# <pre style="max-width: 626px;">

Dan O'Meara

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

[diff] [blame]

806

# PATCH

807

# https://ml.googleapis.com/v1/{name=projects/*/models/*/versions/*}?update_mask=autoScaling.minNodes

808

# -d @./update_body.json

809

# </pre>

810

},

Bu Sun Kim

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

[diff] [blame]

811

"createTime": "A String", # Output only. The time the version was created.

Bu Sun Kim

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

[diff] [blame]

812

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

813

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

814

# are arbitrary strings that you supply.

815

# For more information, see the documentation on

816

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

817

"a_key": "A String",

818

},

Bu Sun Kim

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

[diff] [blame]

819

"requestLoggingConfig": { # Configuration for logging request-response pairs to a BigQuery table. # Optional. *Only* specify this field in a

Dan O'Meara

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

[diff] [blame]

820

# projects.models.versions.patch

821

# request. Specifying it in a

822

# projects.models.versions.create

823

# request has no effect.

824

#

825

# Configures the request-response pair logging on predictions from this

826

# Version.

827

# Online prediction requests to a model version and the responses to these

828

# requests are converted to raw strings and saved to the specified BigQuery

829

# table. Logging is constrained by [BigQuery quotas and

830

# limits](/bigquery/quotas). If your project exceeds BigQuery quotas or limits,

831

# AI Platform Prediction does not log request-response pairs, but it continues

832

# to serve predictions.

833

#

834

# If you are using [continuous

835

# evaluation](/ml-engine/docs/continuous-evaluation/), you do not need to

836

# specify this configuration manually. Setting up continuous evaluation

837

# automatically enables logging of request-response pairs.

Bu Sun Kim

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

[diff] [blame]

838

"bigqueryTableName": "A String", # Required. Fully qualified BigQuery table name in the following format:

839

# "<var>project_id</var>.<var>dataset_name</var>.<var>table_name</var>"

Dan O'Meara

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

[diff] [blame]

840

#

Bu Sun Kim

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

[diff] [blame]

841

# The specified table must already exist, and the "Cloud ML Service Agent"

Dan O'Meara

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

[diff] [blame]

842

# for your project must have permission to write to it. The table must have

843

# the following [schema](/bigquery/docs/schemas):

844

#

845

# <table>

Bu Sun Kim

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

[diff] [blame]

846

# <tr><th>Field name</th><th style="display: table-cell">Type</th>

847

# <th style="display: table-cell">Mode</th></tr>

Dan O'Meara

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

[diff] [blame]

848

# <tr><td>model</td><td>STRING</td><td>REQUIRED</td></tr>

849

# <tr><td>model_version</td><td>STRING</td><td>REQUIRED</td></tr>

850

# <tr><td>time</td><td>TIMESTAMP</td><td>REQUIRED</td></tr>

851

# <tr><td>raw_data</td><td>STRING</td><td>REQUIRED</td></tr>

852

# <tr><td>raw_prediction</td><td>STRING</td><td>NULLABLE</td></tr>

853

# <tr><td>groundtruth</td><td>STRING</td><td>NULLABLE</td></tr>

854

# </table>

Bu Sun Kim

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

[diff] [blame]

855

"samplingPercentage": 3.14, # Percentage of requests to be logged, expressed as a fraction from 0 to 1.

856

# For example, if you want to log 10% of requests, enter `0.1`. The sampling

857

# window is the lifetime of the model version. Defaults to 0.

858

},

859

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

860

"machineType": "A String", # Optional. The type of machine on which to serve the model. Currently only

861

# applies to online prediction service. If this field is not specified, it

862

# defaults to `mls1-c1-m2`.

863

#

864

# Online prediction supports the following machine types:

#

# * `mls1-c1-m2`

# * `mls1-c4-m2`

# * `n1-standard-2`

# * `n1-standard-4`

# * `n1-standard-8`

# * `n1-standard-16`

# * `n1-standard-32`

# * `n1-highmem-2`

# * `n1-highmem-4`

# * `n1-highmem-8`

# * `n1-highmem-16`

# * `n1-highmem-32`

# * `n1-highcpu-2`

# * `n1-highcpu-4`

# * `n1-highcpu-8`

# * `n1-highcpu-16`

# * `n1-highcpu-32`

#

# `mls1-c1-m2` is generally available. All other machine types are available

885

# in beta. Learn more about the [differences between machine

886

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

887

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

888

#

889

# For more information, see the

890

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

891

# [how to manage runtime versions](/ml-engine/docs/versioning).

892

"description": "A String", # Optional. The description specified for the version when it was created.

893

"framework": "A String", # Optional. The machine learning framework AI Platform uses to train

894

# this version of the model. Valid values are `TENSORFLOW`, `SCIKIT_LEARN`,

895

# `XGBOOST`. If you do not specify a framework, AI Platform

896

# will analyze files in the deployment_uri to determine a framework. If you

897

# choose `SCIKIT_LEARN` or `XGBOOST`, you must also set the runtime version

898

# of the model to 1.4 or greater.

899

#

900

# Do **not** specify a framework if you're deploying a [custom

901

# prediction routine](/ml-engine/docs/tensorflow/custom-prediction-routines).

902

#

903

# If you specify a [Compute Engine (N1) machine

904

# type](/ml-engine/docs/machine-types-online-prediction) in the

905

# `machineType` field, you must specify `TENSORFLOW`

906

# for the framework.

907

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

908

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

909

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

910

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

911

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

912

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

913

# ensure that their change will be applied to the model as intended.

Sai Cheemalapati

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

[diff] [blame]

}</pre>

</div>

Bu Sun Kim

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

[diff] [blame]

918

<code class="details" id="list">list(parent, pageToken=None, pageSize=None, filter=None, x__xgafv=None)</code>

Sai Cheemalapati

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

[diff] [blame]

919

<pre>Gets basic information about all the versions of a model.

920

Bu Sun Kim

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

[diff] [blame]

921

If you expect that a model has many versions, or if you need to handle

Sai Cheemalapati

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

[diff] [blame]

922

only a limited number of results at a time, you can request that the list

Bu Sun Kim

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

[diff] [blame]

923

be retrieved in batches (called pages).

924

925

If there are no versions that match the request parameters, the list

926

request returns an empty response body: {}.

Sai Cheemalapati

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

[diff] [blame]

927

928

Args:

Bu Sun Kim

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

[diff] [blame]

929

parent: string, Required. The name of the model for which to list the version. (required)

Sai Cheemalapati

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

[diff] [blame]

930

pageToken: string, Optional. A page token to request the next page of results.

931

932

You get the token from the `next_page_token` field of the response from

933

the previous call.

Bu Sun Kim

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

[diff] [blame]

934

pageSize: integer, Optional. The number of versions to retrieve per "page" of results. If

Sai Cheemalapati

4ba8c23

2017-06-06 18:46:08 -0400

[diff] [blame]

935

there are more remaining results than this number, the response message

936

will contain a valid value in the `next_page_token` field.

937

938

The default value is 20, and the maximum page size is 100.

Bu Sun Kim

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

[diff] [blame]

939

filter: string, Optional. Specifies the subset of versions to retrieve.

Bu Sun Kim

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

[diff] [blame]

940

x__xgafv: string, V1 error format.

941

Allowed values

942

1 - v1 error format

943

2 - v2 error format

Sai Cheemalapati

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

[diff] [blame]

944

945

Returns:

946

An object of the form:

947

948

{ # Response message for the ListVersions method.

Bu Sun Kim

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

[diff] [blame]

949

"nextPageToken": "A String", # Optional. Pass this token as the `page_token` field of the request for a

Sai Cheemalapati

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

[diff] [blame]

950

# subsequent call.

Bu Sun Kim

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

[diff] [blame]

951

"versions": [ # The list of versions.

Sai Cheemalapati

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

[diff] [blame]

952

{ # Represents a version of the model.

953

#

954

# Each version is a trained model deployed in the cloud, ready to handle

955

# prediction requests. A model can have multiple versions. You can get

956

# information about all of the versions of a given model by calling

Dan O'Meara

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

[diff] [blame]

957

# projects.models.versions.list.

Bu Sun Kim

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

[diff] [blame]

958

"acceleratorConfig": { # Represents a hardware accelerator request config. # Optional. Accelerator config for using GPUs for online prediction (beta).

959

# Only specify this field if you have specified a Compute Engine (N1) machine

960

# type in the `machineType` field. Learn more about [using GPUs for online

961

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

962

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

963

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

964

# [accelerators for online

965

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

966

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

967

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

968

},

969

"isDefault": True or False, # Output only. If true, this version will be used to handle prediction

970

# requests that do not specify a version.

971

#

972

# You can change the default version by calling

973

# projects.methods.versions.setDefault.

Bu Sun Kim

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

[diff] [blame]

974

"manualScaling": { # Options for manually scaling a model. # Manually select the number of nodes to use for serving the

975

# model. You should generally use `auto_scaling` with an appropriate

976

# `min_nodes` instead, but this option is available if you want more

977

# predictable billing. Beware that latency and error rates will increase

978

# if the traffic exceeds that capability of the system to serve it based

979

# on the selected number of nodes.

980

"nodes": 42, # The number of nodes to allocate for this model. These nodes are always up,

981

# starting from the time the model is deployed, so the cost of operating

982

# this model will be proportional to `nodes` * number of hours since

983

# last billing cycle plus the cost for each prediction performed.

Dan O'Meara

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

[diff] [blame]

984

},

Bu Sun Kim

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

[diff] [blame]

985

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

Bu Sun Kim

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

[diff] [blame]

986

"name": "A String", # Required. The name specified for the version when it was created.

987

#

988

# The version name must be unique within the model it is created in.

989

"serviceAccount": "A String", # Optional. Specifies the service account for resource access control.

990

"pythonVersion": "A String", # Required. The version of Python used in prediction.

991

#

992

# The following Python versions are available:

993

#

994

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

995

# later.

996

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

997

# from '1.4' to '1.14'.

998

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

999

# earlier.

1000

#

1001

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

1002

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

1003

"lastUseTime": "A String", # Output only. The time the version was last used for prediction.

1004

"predictionClass": "A String", # Optional. The fully qualified name

Dan O'Meara

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

[diff] [blame]

1005

# (<var>module_name</var>.<var>class_name</var>) of a class that implements

Bu Sun Kim

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

[diff] [blame]

1006

# the Predictor interface described in this reference field. The module

1007

# containing this class should be included in a package provided to the

1008

# [`packageUris` field](#Version.FIELDS.package_uris).

Sai Cheemalapati

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

[diff] [blame]

1009

#

Bu Sun Kim

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

[diff] [blame]

1010

# Specify this field if and only if you are deploying a [custom prediction

1011

# routine (beta)](/ml-engine/docs/tensorflow/custom-prediction-routines).

1012

# If you specify this field, you must set

Dan O'Meara

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

[diff] [blame]

1013

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater and

1014

# you must set `machineType` to a [legacy (MLS1)

1015

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

Bu Sun Kim

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

[diff] [blame]

1016

#

1017

# The following code sample provides the Predictor interface:

1018

#

Bu Sun Kim

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

[diff] [blame]

1019

# <pre style="max-width: 626px;">

Bu Sun Kim

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

[diff] [blame]

1020

# class Predictor(object):

Bu Sun Kim

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

[diff] [blame]

1021

# """Interface for constructing custom predictors."""

Bu Sun Kim

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

[diff] [blame]

1022

#

1023

# def predict(self, instances, **kwargs):

Bu Sun Kim

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

[diff] [blame]

1024

# """Performs custom prediction.

Bu Sun Kim

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

[diff] [blame]

1025

#

1026

# Instances are the decoded values from the request. They have already

1027

# been deserialized from JSON.

1028

#

1029

# Args:

1030

# instances: A list of prediction input instances.

1031

# **kwargs: A dictionary of keyword args provided as additional

1032

# fields on the predict request body.

1033

#

1034

# Returns:

1035

# A list of outputs containing the prediction results. This list must

1036

# be JSON serializable.

Bu Sun Kim

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

[diff] [blame]

1037

# """

Bu Sun Kim

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

[diff] [blame]

1038

# raise NotImplementedError()

1039

#

1040

# @classmethod

1041

# def from_path(cls, model_dir):

Bu Sun Kim

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

[diff] [blame]

1042

# """Creates an instance of Predictor using the given path.

Bu Sun Kim

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

[diff] [blame]

1043

#

1044

# Loading of the predictor should be done in this method.

1045

#

1046

# Args:

1047

# model_dir: The local directory that contains the exported model

1048

# file along with any additional files uploaded when creating the

# version resource.

#

# Returns:

# An instance implementing this Predictor class.

Bu Sun Kim

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

[diff] [blame]

1053

# """

Bu Sun Kim

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

[diff] [blame]

1054

# raise NotImplementedError()

Dan O'Meara

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

[diff] [blame]

1055

# </pre>

Bu Sun Kim

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

[diff] [blame]

1056

#

1057

# Learn more about [the Predictor interface and custom prediction

1058

# routines](/ml-engine/docs/tensorflow/custom-prediction-routines).

Bu Sun Kim

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

[diff] [blame]

1059

"deploymentUri": "A String", # Required. The Cloud Storage location of the trained model used to

1060

# create the version. See the

1061

# [guide to model

1062

# deployment](/ml-engine/docs/tensorflow/deploying-models) for more

1063

# information.

1064

#

1065

# When passing Version to

1066

# projects.models.versions.create

1067

# the model service uses the specified location as the source of the model.

1068

# Once deployed, the model version is hosted by the prediction service, so

1069

# this location is useful only as a historical record.

1070

# The total number of model files can't exceed 1000.

Bu Sun Kim

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

[diff] [blame]

1071

"packageUris": [ # Optional. Cloud Storage paths (`gs://…`) of packages for [custom

Bu Sun Kim

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

[diff] [blame]

1072

# prediction routines](/ml-engine/docs/tensorflow/custom-prediction-routines)

1073

# or [scikit-learn pipelines with custom

1074

# code](/ml-engine/docs/scikit/exporting-for-prediction#custom-pipeline-code).

1075

#

1076

# For a custom prediction routine, one of these packages must contain your

1077

# Predictor class (see

1078

# [`predictionClass`](#Version.FIELDS.prediction_class)). Additionally,

1079

# include any dependencies used by your Predictor or scikit-learn pipeline

1080

# uses that are not already included in your selected [runtime

1081

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

1082

#

1083

# If you specify this field, you must also set

1084

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater.

Bu Sun Kim

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

[diff] [blame]

1085

"A String",

Bu Sun Kim

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

[diff] [blame]

1086

],

Bu Sun Kim

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

[diff] [blame]

1087

"explanationConfig": { # Message holding configuration options for explaining model predictions. # Optional. Configures explainability features on the model's version.

1088

# Some explanation features require additional metadata to be loaded

1089

# as part of the model payload.

1090

# There are two feature attribution methods supported for TensorFlow models:

1091

# integrated gradients and sampled Shapley.

1092

# [Learn more about feature

1093

# attributions.](/ai-platform/prediction/docs/ai-explanations/overview)

1094

"integratedGradientsAttribution": { # Attributes credit by computing the Aumann-Shapley value taking advantage # Attributes credit by computing the Aumann-Shapley value taking advantage

1095

# of the model's fully differentiable structure. Refer to this paper for

1096

# more details: http://proceedings.mlr.press/v70/sundararajan17a.html

1097

# of the model's fully differentiable structure. Refer to this paper for

1098

# more details: https://arxiv.org/abs/1703.01365

1099

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

1100

# A good value to start is 50 and gradually increase until the

1101

# sum to diff property is met within the desired error range.

1102

},

1103

"sampledShapleyAttribution": { # An attribution method that approximates Shapley values for features that # An attribution method that approximates Shapley values for features that

1104

# contribute to the label being predicted. A sampling strategy is used to

1105

# approximate the value rather than considering all subsets of features.

1106

# contribute to the label being predicted. A sampling strategy is used to

1107

# approximate the value rather than considering all subsets of features.

1108

"numPaths": 42, # The number of feature permutations to consider when approximating the

1109

# Shapley values.

1110

},

1111

"xraiAttribution": { # Attributes credit by computing the XRAI taking advantage # Attributes credit by computing the XRAI taking advantage

1112

# of the model's fully differentiable structure. Refer to this paper for

1113

# more details: https://arxiv.org/abs/1906.02825

1114

# Currently only implemented for models with natural image inputs.

1115

# of the model's fully differentiable structure. Refer to this paper for

1116

# more details: https://arxiv.org/abs/1906.02825

1117

# Currently only implemented for models with natural image inputs.

1118

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

1119

# A good value to start is 50 and gradually increase until the

1120

# sum to diff property is met within the desired error range.

1121

},

1122

},

Bu Sun Kim

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

[diff] [blame]

1123

"autoScaling": { # Options for automatically scaling a model. # Automatically scale the number of nodes used to serve the model in

Dan O'Meara

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

[diff] [blame]

1124

# response to increases and decreases in traffic. Care should be

Bu Sun Kim

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

[diff] [blame]

1125

# taken to ramp up traffic according to the model's ability to scale

Dan O'Meara

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

[diff] [blame]

1126

# or you will start seeing increases in latency and 429 response codes.

1127

#

1128

# Note that you cannot use AutoScaling if your version uses

1129

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use specify

1130

# `manual_scaling`.

Bu Sun Kim

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

[diff] [blame]

1131

"minNodes": 42, # Optional. The minimum number of nodes to allocate for this model. These

Dan O'Meara

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

[diff] [blame]

1132

# nodes are always up, starting from the time the model is deployed.

1133

# Therefore, the cost of operating this model will be at least

1134

# `rate` * `min_nodes` * number of hours since last billing cycle,

1135

# where `rate` is the cost per node-hour as documented in the

1136

# [pricing guide](/ml-engine/docs/pricing),

1137

# even if no predictions are performed. There is additional cost for each

1138

# prediction performed.

1139

#

1140

# Unlike manual scaling, if the load gets too heavy for the nodes

1141

# that are up, the service will automatically add nodes to handle the

1142

# increased load as well as scale back as traffic drops, always maintaining

1143

# at least `min_nodes`. You will be charged for the time in which additional

1144

# nodes are used.

1145

#

1146

# If `min_nodes` is not specified and AutoScaling is used with a [legacy

1147

# (MLS1) machine type](/ml-engine/docs/machine-types-online-prediction),

1148

# `min_nodes` defaults to 0, in which case, when traffic to a model stops

1149

# (and after a cool-down period), nodes will be shut down and no charges will

1150

# be incurred until traffic to the model resumes.

1151

#

1152

# If `min_nodes` is not specified and AutoScaling is used with a [Compute

1153

# Engine (N1) machine type](/ml-engine/docs/machine-types-online-prediction),

1154

# `min_nodes` defaults to 1. `min_nodes` must be at least 1 for use with a

1155

# Compute Engine machine type.

1156

#

1157

# Note that you cannot use AutoScaling if your version uses

1158

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use

1159

# ManualScaling.

1160

#

1161

# You can set `min_nodes` when creating the model version, and you can also

1162

# update `min_nodes` for an existing version:

1163

# <pre>

1164

# update_body.json:

1165

# {

Bu Sun Kim

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

[diff] [blame]

1166

# 'autoScaling': {

1167

# 'minNodes': 5

Dan O'Meara

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

[diff] [blame]

# }

# }

# </pre>

# HTTP request:

Bu Sun Kim

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

[diff] [blame]

1172

# <pre style="max-width: 626px;">

Dan O'Meara

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

[diff] [blame]

1173

# PATCH

1174

# https://ml.googleapis.com/v1/{name=projects/*/models/*/versions/*}?update_mask=autoScaling.minNodes

1175

# -d @./update_body.json

1176

# </pre>

1177

},

Bu Sun Kim

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

[diff] [blame]

1178

"createTime": "A String", # Output only. The time the version was created.

Bu Sun Kim

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

[diff] [blame]

1179

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

1180

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

1181

# are arbitrary strings that you supply.

1182

# For more information, see the documentation on

1183

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

1184

"a_key": "A String",

1185

},

Bu Sun Kim

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

[diff] [blame]

1186

"requestLoggingConfig": { # Configuration for logging request-response pairs to a BigQuery table. # Optional. *Only* specify this field in a

Dan O'Meara

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

[diff] [blame]

1187

# projects.models.versions.patch

1188

# request. Specifying it in a

1189

# projects.models.versions.create

1190

# request has no effect.

1191

#

1192

# Configures the request-response pair logging on predictions from this

1193

# Version.

1194

# Online prediction requests to a model version and the responses to these

1195

# requests are converted to raw strings and saved to the specified BigQuery

1196

# table. Logging is constrained by [BigQuery quotas and

1197

# limits](/bigquery/quotas). If your project exceeds BigQuery quotas or limits,

1198

# AI Platform Prediction does not log request-response pairs, but it continues

1199

# to serve predictions.

1200

#

1201

# If you are using [continuous

1202

# evaluation](/ml-engine/docs/continuous-evaluation/), you do not need to

1203

# specify this configuration manually. Setting up continuous evaluation

1204

# automatically enables logging of request-response pairs.

Bu Sun Kim

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

[diff] [blame]

1205

"bigqueryTableName": "A String", # Required. Fully qualified BigQuery table name in the following format:

1206

# "<var>project_id</var>.<var>dataset_name</var>.<var>table_name</var>"

Dan O'Meara

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

[diff] [blame]

1207

#

Bu Sun Kim

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

[diff] [blame]

1208

# The specified table must already exist, and the "Cloud ML Service Agent"

Dan O'Meara

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

[diff] [blame]

1209

# for your project must have permission to write to it. The table must have

1210

# the following [schema](/bigquery/docs/schemas):

1211

#

1212

# <table>

Bu Sun Kim

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

[diff] [blame]

1213

# <tr><th>Field name</th><th style="display: table-cell">Type</th>

1214

# <th style="display: table-cell">Mode</th></tr>

Dan O'Meara

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

[diff] [blame]

1215

# <tr><td>model</td><td>STRING</td><td>REQUIRED</td></tr>

1216

# <tr><td>model_version</td><td>STRING</td><td>REQUIRED</td></tr>

1217

# <tr><td>time</td><td>TIMESTAMP</td><td>REQUIRED</td></tr>

1218

# <tr><td>raw_data</td><td>STRING</td><td>REQUIRED</td></tr>

1219

# <tr><td>raw_prediction</td><td>STRING</td><td>NULLABLE</td></tr>

1220

# <tr><td>groundtruth</td><td>STRING</td><td>NULLABLE</td></tr>

1221

# </table>

Bu Sun Kim

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

[diff] [blame]

1222

"samplingPercentage": 3.14, # Percentage of requests to be logged, expressed as a fraction from 0 to 1.

1223

# For example, if you want to log 10% of requests, enter `0.1`. The sampling

1224

# window is the lifetime of the model version. Defaults to 0.

1225

},

1226

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

1227

"machineType": "A String", # Optional. The type of machine on which to serve the model. Currently only

1228

# applies to online prediction service. If this field is not specified, it

1229

# defaults to `mls1-c1-m2`.

1230

#

1231

# Online prediction supports the following machine types:

#

# * `mls1-c1-m2`

# * `mls1-c4-m2`

# * `n1-standard-2`

# * `n1-standard-4`

# * `n1-standard-8`

# * `n1-standard-16`

# * `n1-standard-32`

# * `n1-highmem-2`

# * `n1-highmem-4`

# * `n1-highmem-8`

# * `n1-highmem-16`

# * `n1-highmem-32`

# * `n1-highcpu-2`

# * `n1-highcpu-4`

# * `n1-highcpu-8`

# * `n1-highcpu-16`

# * `n1-highcpu-32`

#

# `mls1-c1-m2` is generally available. All other machine types are available

1252

# in beta. Learn more about the [differences between machine

1253

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

1254

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

1255

#

1256

# For more information, see the

1257

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

1258

# [how to manage runtime versions](/ml-engine/docs/versioning).

1259

"description": "A String", # Optional. The description specified for the version when it was created.

1260

"framework": "A String", # Optional. The machine learning framework AI Platform uses to train

1261

# this version of the model. Valid values are `TENSORFLOW`, `SCIKIT_LEARN`,

1262

# `XGBOOST`. If you do not specify a framework, AI Platform

1263

# will analyze files in the deployment_uri to determine a framework. If you

1264

# choose `SCIKIT_LEARN` or `XGBOOST`, you must also set the runtime version

1265

# of the model to 1.4 or greater.

1266

#

1267

# Do **not** specify a framework if you're deploying a [custom

1268

# prediction routine](/ml-engine/docs/tensorflow/custom-prediction-routines).

1269

#

1270

# If you specify a [Compute Engine (N1) machine

1271

# type](/ml-engine/docs/machine-types-online-prediction) in the

1272

# `machineType` field, you must specify `TENSORFLOW`

1273

# for the framework.

1274

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

1275

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

1276

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

1277

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

1278

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

1279

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

1280

# ensure that their change will be applied to the model as intended.

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>

1288

<pre>Retrieves the next page of results.

1289

1290

Args:

1291

previous_request: The request for the previous page. (required)

1292

previous_response: The response from the request for the previous page. (required)

1293

1294

Returns:

Bu Sun Kim

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

[diff] [blame]

1295

A request object that you can call 'execute()' on to request the next

Sai Cheemalapati

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

[diff] [blame]

1296

page. Returns None if there are no more items in the collection.

</pre>

</div>

Dan O'Meara

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

[diff] [blame]

1301

<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]

1302

<pre>Updates the specified Version resource.

1303

Dan O'Meara

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

[diff] [blame]

1304

Currently the only update-able fields are `description`,

1305

`requestLoggingConfig`, `autoScaling.minNodes`, and `manualScaling.nodes`.

Bu Sun Kim

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

[diff] [blame]

1306

1307

Args:

1308

name: string, Required. The name of the model. (required)

Dan O'Meara

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

[diff] [blame]

1309

body: object, The request body.

Bu Sun Kim

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

[diff] [blame]

1310

The object takes the form of:

1311

1312

{ # Represents a version of the model.

1313

#

1314

# Each version is a trained model deployed in the cloud, ready to handle

1315

# prediction requests. A model can have multiple versions. You can get

1316

# information about all of the versions of a given model by calling

Dan O'Meara

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

[diff] [blame]

1317

# projects.models.versions.list.

Bu Sun Kim

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

[diff] [blame]

1318

"acceleratorConfig": { # Represents a hardware accelerator request config. # Optional. Accelerator config for using GPUs for online prediction (beta).

1319

# Only specify this field if you have specified a Compute Engine (N1) machine

1320

# type in the `machineType` field. Learn more about [using GPUs for online

1321

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

1322

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

1323

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

1324

# [accelerators for online

1325

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

1326

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

1327

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

1328

},

1329

"isDefault": True or False, # Output only. If true, this version will be used to handle prediction

1330

# requests that do not specify a version.

1331

#

1332

# You can change the default version by calling

1333

# projects.methods.versions.setDefault.

Bu Sun Kim

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

[diff] [blame]

1334

"manualScaling": { # Options for manually scaling a model. # Manually select the number of nodes to use for serving the

1335

# model. You should generally use `auto_scaling` with an appropriate

1336

# `min_nodes` instead, but this option is available if you want more

1337

# predictable billing. Beware that latency and error rates will increase

1338

# if the traffic exceeds that capability of the system to serve it based

1339

# on the selected number of nodes.

1340

"nodes": 42, # The number of nodes to allocate for this model. These nodes are always up,

1341

# starting from the time the model is deployed, so the cost of operating

1342

# this model will be proportional to `nodes` * number of hours since

1343

# last billing cycle plus the cost for each prediction performed.

Dan O'Meara

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

[diff] [blame]

1344

},

Bu Sun Kim

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

[diff] [blame]

1345

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

Bu Sun Kim

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

[diff] [blame]

1346

"name": "A String", # Required. The name specified for the version when it was created.

1347

#

1348

# The version name must be unique within the model it is created in.

1349

"serviceAccount": "A String", # Optional. Specifies the service account for resource access control.

1350

"pythonVersion": "A String", # Required. The version of Python used in prediction.

1351

#

1352

# The following Python versions are available:

1353

#

1354

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

1355

# later.

1356

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

1357

# from '1.4' to '1.14'.

1358

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

1359

# earlier.

1360

#

1361

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

1362

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

1363

"lastUseTime": "A String", # Output only. The time the version was last used for prediction.

1364

"predictionClass": "A String", # Optional. The fully qualified name

Dan O'Meara

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

[diff] [blame]

1365

# (<var>module_name</var>.<var>class_name</var>) of a class that implements

Bu Sun Kim

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

[diff] [blame]

1366

# the Predictor interface described in this reference field. The module

1367

# containing this class should be included in a package provided to the

1368

# [`packageUris` field](#Version.FIELDS.package_uris).

1369

#

1370

# Specify this field if and only if you are deploying a [custom prediction

1371

# routine (beta)](/ml-engine/docs/tensorflow/custom-prediction-routines).

1372

# If you specify this field, you must set

Dan O'Meara

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

[diff] [blame]

1373

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater and

1374

# you must set `machineType` to a [legacy (MLS1)

1375

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

Bu Sun Kim

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

[diff] [blame]

1376

#

1377

# The following code sample provides the Predictor interface:

1378

#

Bu Sun Kim

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

[diff] [blame]

1379

# <pre style="max-width: 626px;">

Bu Sun Kim

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

[diff] [blame]

1380

# class Predictor(object):

Bu Sun Kim

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

[diff] [blame]

1381

# """Interface for constructing custom predictors."""

Bu Sun Kim

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

[diff] [blame]

1382

#

1383

# def predict(self, instances, **kwargs):

Bu Sun Kim

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

[diff] [blame]

1384

# """Performs custom prediction.

Bu Sun Kim

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

[diff] [blame]

1385

#

1386

# Instances are the decoded values from the request. They have already

1387

# been deserialized from JSON.

1388

#

1389

# Args:

1390

# instances: A list of prediction input instances.

1391

# **kwargs: A dictionary of keyword args provided as additional

1392

# fields on the predict request body.

1393

#

1394

# Returns:

1395

# A list of outputs containing the prediction results. This list must

1396

# be JSON serializable.

Bu Sun Kim

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

[diff] [blame]

1397

# """

Bu Sun Kim

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

[diff] [blame]

1398

# raise NotImplementedError()

1399

#

1400

# @classmethod

1401

# def from_path(cls, model_dir):

Bu Sun Kim

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

[diff] [blame]

1402

# """Creates an instance of Predictor using the given path.

Bu Sun Kim

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

[diff] [blame]

1403

#

1404

# Loading of the predictor should be done in this method.

1405

#

1406

# Args:

1407

# model_dir: The local directory that contains the exported model

1408

# file along with any additional files uploaded when creating the

# version resource.

#

# Returns:

# An instance implementing this Predictor class.

Bu Sun Kim

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

[diff] [blame]

1413

# """

Bu Sun Kim

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

[diff] [blame]

1414

# raise NotImplementedError()

Dan O'Meara

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

[diff] [blame]

1415

# </pre>

Bu Sun Kim

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

[diff] [blame]

1416

#

1417

# Learn more about [the Predictor interface and custom prediction

1418

# routines](/ml-engine/docs/tensorflow/custom-prediction-routines).

Bu Sun Kim

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

[diff] [blame]

1419

"deploymentUri": "A String", # Required. The Cloud Storage location of the trained model used to

1420

# create the version. See the

1421

# [guide to model

1422

# deployment](/ml-engine/docs/tensorflow/deploying-models) for more

1423

# information.

1424

#

1425

# When passing Version to

1426

# projects.models.versions.create

1427

# the model service uses the specified location as the source of the model.

1428

# Once deployed, the model version is hosted by the prediction service, so

1429

# this location is useful only as a historical record.

1430

# The total number of model files can't exceed 1000.

Bu Sun Kim

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

[diff] [blame]

1431

"packageUris": [ # Optional. Cloud Storage paths (`gs://…`) of packages for [custom

Bu Sun Kim

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

[diff] [blame]

1432

# prediction routines](/ml-engine/docs/tensorflow/custom-prediction-routines)

1433

# or [scikit-learn pipelines with custom

1434

# code](/ml-engine/docs/scikit/exporting-for-prediction#custom-pipeline-code).

1435

#

1436

# For a custom prediction routine, one of these packages must contain your

1437

# Predictor class (see

1438

# [`predictionClass`](#Version.FIELDS.prediction_class)). Additionally,

1439

# include any dependencies used by your Predictor or scikit-learn pipeline

1440

# uses that are not already included in your selected [runtime

1441

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

1442

#

1443

# If you specify this field, you must also set

1444

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater.

Bu Sun Kim

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

[diff] [blame]

1445

"A String",

Bu Sun Kim

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

[diff] [blame]

1446

],

Bu Sun Kim

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

[diff] [blame]

1447

"explanationConfig": { # Message holding configuration options for explaining model predictions. # Optional. Configures explainability features on the model's version.

1448

# Some explanation features require additional metadata to be loaded

1449

# as part of the model payload.

1450

# There are two feature attribution methods supported for TensorFlow models:

1451

# integrated gradients and sampled Shapley.

1452

# [Learn more about feature

1453

# attributions.](/ai-platform/prediction/docs/ai-explanations/overview)

1454

"integratedGradientsAttribution": { # Attributes credit by computing the Aumann-Shapley value taking advantage # Attributes credit by computing the Aumann-Shapley value taking advantage

1455

# of the model's fully differentiable structure. Refer to this paper for

1456

# more details: http://proceedings.mlr.press/v70/sundararajan17a.html

1457

# of the model's fully differentiable structure. Refer to this paper for

1458

# more details: https://arxiv.org/abs/1703.01365

1459

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

1460

# A good value to start is 50 and gradually increase until the

1461

# sum to diff property is met within the desired error range.

1462

},

1463

"sampledShapleyAttribution": { # An attribution method that approximates Shapley values for features that # An attribution method that approximates Shapley values for features that

1464

# contribute to the label being predicted. A sampling strategy is used to

1465

# approximate the value rather than considering all subsets of features.

1466

# contribute to the label being predicted. A sampling strategy is used to

1467

# approximate the value rather than considering all subsets of features.

1468

"numPaths": 42, # The number of feature permutations to consider when approximating the

1469

# Shapley values.

1470

},

1471

"xraiAttribution": { # Attributes credit by computing the XRAI taking advantage # Attributes credit by computing the XRAI taking advantage

1472

# of the model's fully differentiable structure. Refer to this paper for

1473

# more details: https://arxiv.org/abs/1906.02825

1474

# Currently only implemented for models with natural image inputs.

1475

# of the model's fully differentiable structure. Refer to this paper for

1476

# more details: https://arxiv.org/abs/1906.02825

1477

# Currently only implemented for models with natural image inputs.

1478

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

1479

# A good value to start is 50 and gradually increase until the

1480

# sum to diff property is met within the desired error range.

1481

},

1482

},

Bu Sun Kim

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

[diff] [blame]

1483

"autoScaling": { # Options for automatically scaling a model. # Automatically scale the number of nodes used to serve the model in

Dan O'Meara

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

[diff] [blame]

1484

# response to increases and decreases in traffic. Care should be

Bu Sun Kim

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

[diff] [blame]

1485

# taken to ramp up traffic according to the model's ability to scale

Dan O'Meara

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

[diff] [blame]

1486

# or you will start seeing increases in latency and 429 response codes.

1487

#

1488

# Note that you cannot use AutoScaling if your version uses

1489

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use specify

1490

# `manual_scaling`.

Bu Sun Kim

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

[diff] [blame]

1491

"minNodes": 42, # Optional. The minimum number of nodes to allocate for this model. These

Dan O'Meara

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

[diff] [blame]

1492

# nodes are always up, starting from the time the model is deployed.

1493

# Therefore, the cost of operating this model will be at least

1494

# `rate` * `min_nodes` * number of hours since last billing cycle,

1495

# where `rate` is the cost per node-hour as documented in the

1496

# [pricing guide](/ml-engine/docs/pricing),

1497

# even if no predictions are performed. There is additional cost for each

1498

# prediction performed.

1499

#

1500

# Unlike manual scaling, if the load gets too heavy for the nodes

1501

# that are up, the service will automatically add nodes to handle the

1502

# increased load as well as scale back as traffic drops, always maintaining

1503

# at least `min_nodes`. You will be charged for the time in which additional

1504

# nodes are used.

1505

#

1506

# If `min_nodes` is not specified and AutoScaling is used with a [legacy

1507

# (MLS1) machine type](/ml-engine/docs/machine-types-online-prediction),

1508

# `min_nodes` defaults to 0, in which case, when traffic to a model stops

1509

# (and after a cool-down period), nodes will be shut down and no charges will

1510

# be incurred until traffic to the model resumes.

1511

#

1512

# If `min_nodes` is not specified and AutoScaling is used with a [Compute

1513

# Engine (N1) machine type](/ml-engine/docs/machine-types-online-prediction),

1514

# `min_nodes` defaults to 1. `min_nodes` must be at least 1 for use with a

1515

# Compute Engine machine type.

1516

#

1517

# Note that you cannot use AutoScaling if your version uses

1518

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use

1519

# ManualScaling.

1520

#

1521

# You can set `min_nodes` when creating the model version, and you can also

1522

# update `min_nodes` for an existing version:

1523

# <pre>

1524

# update_body.json:

1525

# {

Bu Sun Kim

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

[diff] [blame]

1526

# 'autoScaling': {

1527

# 'minNodes': 5

Dan O'Meara

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

[diff] [blame]

# }

# }

# </pre>

# HTTP request:

Bu Sun Kim

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

[diff] [blame]

1532

# <pre style="max-width: 626px;">

Dan O'Meara

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

[diff] [blame]

1533

# PATCH

1534

# https://ml.googleapis.com/v1/{name=projects/*/models/*/versions/*}?update_mask=autoScaling.minNodes

1535

# -d @./update_body.json

1536

# </pre>

1537

},

Bu Sun Kim

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

[diff] [blame]

1538

"createTime": "A String", # Output only. The time the version was created.

Bu Sun Kim

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

[diff] [blame]

1539

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

1540

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

1541

# are arbitrary strings that you supply.

1542

# For more information, see the documentation on

1543

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

1544

"a_key": "A String",

1545

},

Bu Sun Kim

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

[diff] [blame]

1546

"requestLoggingConfig": { # Configuration for logging request-response pairs to a BigQuery table. # Optional. *Only* specify this field in a

Dan O'Meara

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

[diff] [blame]

1547

# projects.models.versions.patch

1548

# request. Specifying it in a

1549

# projects.models.versions.create

1550

# request has no effect.

1551

#

1552

# Configures the request-response pair logging on predictions from this

1553

# Version.

1554

# Online prediction requests to a model version and the responses to these

1555

# requests are converted to raw strings and saved to the specified BigQuery

1556

# table. Logging is constrained by [BigQuery quotas and

1557

# limits](/bigquery/quotas). If your project exceeds BigQuery quotas or limits,

1558

# AI Platform Prediction does not log request-response pairs, but it continues

1559

# to serve predictions.

1560

#

1561

# If you are using [continuous

1562

# evaluation](/ml-engine/docs/continuous-evaluation/), you do not need to

1563

# specify this configuration manually. Setting up continuous evaluation

1564

# automatically enables logging of request-response pairs.

Bu Sun Kim

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

[diff] [blame]

1565

"bigqueryTableName": "A String", # Required. Fully qualified BigQuery table name in the following format:

1566

# "<var>project_id</var>.<var>dataset_name</var>.<var>table_name</var>"

Dan O'Meara

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

[diff] [blame]

1567

#

Bu Sun Kim

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

[diff] [blame]

1568

# The specified table must already exist, and the "Cloud ML Service Agent"

Dan O'Meara

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

[diff] [blame]

1569

# for your project must have permission to write to it. The table must have

1570

# the following [schema](/bigquery/docs/schemas):

1571

#

1572

# <table>

Bu Sun Kim

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

[diff] [blame]

1573

# <tr><th>Field name</th><th style="display: table-cell">Type</th>

1574

# <th style="display: table-cell">Mode</th></tr>

Dan O'Meara

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

[diff] [blame]

1575

# <tr><td>model</td><td>STRING</td><td>REQUIRED</td></tr>

1576

# <tr><td>model_version</td><td>STRING</td><td>REQUIRED</td></tr>

1577

# <tr><td>time</td><td>TIMESTAMP</td><td>REQUIRED</td></tr>

1578

# <tr><td>raw_data</td><td>STRING</td><td>REQUIRED</td></tr>

1579

# <tr><td>raw_prediction</td><td>STRING</td><td>NULLABLE</td></tr>

1580

# <tr><td>groundtruth</td><td>STRING</td><td>NULLABLE</td></tr>

1581

# </table>

Bu Sun Kim

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

[diff] [blame]

1582

"samplingPercentage": 3.14, # Percentage of requests to be logged, expressed as a fraction from 0 to 1.

1583

# For example, if you want to log 10% of requests, enter `0.1`. The sampling

1584

# window is the lifetime of the model version. Defaults to 0.

1585

},

1586

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

1587

"machineType": "A String", # Optional. The type of machine on which to serve the model. Currently only

1588

# applies to online prediction service. If this field is not specified, it

1589

# defaults to `mls1-c1-m2`.

1590

#

1591

# Online prediction supports the following machine types:

#

# * `mls1-c1-m2`

# * `mls1-c4-m2`

# * `n1-standard-2`

# * `n1-standard-4`

# * `n1-standard-8`

# * `n1-standard-16`

# * `n1-standard-32`

# * `n1-highmem-2`

# * `n1-highmem-4`

# * `n1-highmem-8`

# * `n1-highmem-16`

# * `n1-highmem-32`

# * `n1-highcpu-2`

# * `n1-highcpu-4`

# * `n1-highcpu-8`

# * `n1-highcpu-16`

# * `n1-highcpu-32`

#

# `mls1-c1-m2` is generally available. All other machine types are available

1612

# in beta. Learn more about the [differences between machine

1613

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

1614

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

1615

#

1616

# For more information, see the

1617

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

1618

# [how to manage runtime versions](/ml-engine/docs/versioning).

1619

"description": "A String", # Optional. The description specified for the version when it was created.

1620

"framework": "A String", # Optional. The machine learning framework AI Platform uses to train

1621

# this version of the model. Valid values are `TENSORFLOW`, `SCIKIT_LEARN`,

1622

# `XGBOOST`. If you do not specify a framework, AI Platform

1623

# will analyze files in the deployment_uri to determine a framework. If you

1624

# choose `SCIKIT_LEARN` or `XGBOOST`, you must also set the runtime version

1625

# of the model to 1.4 or greater.

1626

#

1627

# Do **not** specify a framework if you're deploying a [custom

1628

# prediction routine](/ml-engine/docs/tensorflow/custom-prediction-routines).

1629

#

1630

# If you specify a [Compute Engine (N1) machine

1631

# type](/ml-engine/docs/machine-types-online-prediction) in the

1632

# `machineType` field, you must specify `TENSORFLOW`

1633

# for the framework.

1634

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

1635

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

1636

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

1637

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

1638

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

1639

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

1640

# ensure that their change will be applied to the model as intended.

Bu Sun Kim

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

[diff] [blame]

1641

}

1642

1643

updateMask: string, Required. Specifies the path, relative to `Version`, of the field to

1644

update. Must be present and non-empty.

1645

Bu Sun Kim

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

[diff] [blame]

1646

For example, to change the description of a version to "foo", the

Bu Sun Kim

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

[diff] [blame]

1647

`update_mask` parameter would be specified as `description`, and the

1648

`PATCH` request body would specify the new value, as follows:

Bu Sun Kim

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

[diff] [blame]

1649

Dan O'Meara

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

[diff] [blame]

1650

```

1651

{

Bu Sun Kim

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

[diff] [blame]

1652

"description": "foo"

Dan O'Meara

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

[diff] [blame]

}

```

Currently the only supported update mask fields are `description`,

1657

`requestLoggingConfig`, `autoScaling.minNodes`, and `manualScaling.nodes`.

1658

However, you can only update `manualScaling.nodes` if the version uses a

1659

[Compute Engine (N1)

1660

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

Bu Sun Kim

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

[diff] [blame]

1661

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

1668

1669

{ # This resource represents a long-running operation that is the result of a

1670

# network API call.

Bu Sun Kim

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

[diff] [blame]

1671

"name": "A String", # The server-assigned name, which is only unique within the same service that

Bu Sun Kim

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

[diff] [blame]

1672

# originally returns it. If you use the default HTTP mapping, the

1673

# `name` should be a resource name ending with `operations/{unique_id}`.

Bu Sun Kim

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

[diff] [blame]

1674

"error": { # The `Status` type defines a logical error model that is suitable for # The error result of the operation in case of failure or cancellation.

1675

# different programming environments, including REST APIs and RPC APIs. It is

1676

# used by [gRPC](https://github.com/grpc). Each `Status` message contains

1677

# three pieces of data: error code, error message, and error details.

1678

#

1679

# You can find out more about this error model and how to work with it in the

1680

# [API Design Guide](https://cloud.google.com/apis/design/errors).

Bu Sun Kim

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

[diff] [blame]

1681

"code": 42, # The status code, which should be an enum value of google.rpc.Code.

1682

"message": "A String", # A developer-facing error message, which should be in English. Any

1683

# user-facing error message should be localized and sent in the

1684

# google.rpc.Status.details field, or localized by the client.

Bu Sun Kim

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

[diff] [blame]

1685

"details": [ # A list of messages that carry the error details. There is a common set of

1686

# message types for APIs to use.

1687

{

1688

"a_key": "", # Properties of the object. Contains field @type with type URL.

1689

},

1690

],

Bu Sun Kim

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

[diff] [blame]

1691

},

1692

"metadata": { # Service-specific metadata associated with the operation. It typically

1693

# contains progress information and common metadata such as create time.

1694

# Some services might not provide such metadata. Any method that returns a

1695

# long-running operation should document the metadata type, if any.

1696

"a_key": "", # Properties of the object. Contains field @type with type URL.

1697

},

1698

"done": True or False, # If the value is `false`, it means the operation is still in progress.

1699

# If `true`, the operation is completed, and either `error` or `response` is

1700

# available.

Bu Sun Kim

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

[diff] [blame]

1701

"response": { # The normal response of the operation in case of success. If the original

1702

# method returns no data on success, such as `Delete`, the response is

1703

# `google.protobuf.Empty`. If the original method is standard

1704

# `Get`/`Create`/`Update`, the response should be the resource. For other

1705

# methods, the response should have the type `XxxResponse`, where `Xxx`

1706

# is the original method name. For example, if the original method name

1707

# is `TakeSnapshot()`, the inferred response type is

1708

# `TakeSnapshotResponse`.

1709

"a_key": "", # Properties of the object. Contains field @type with type URL.

1710

},

Bu Sun Kim

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

[diff] [blame]

}</pre>

</div>

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

Sai Cheemalapati

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

[diff] [blame]

1716

<pre>Designates a version to be the default for the model.

1717

1718

The default version is used for prediction requests made against the model

Bu Sun Kim

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

[diff] [blame]

1719

that don't specify a version.

Sai Cheemalapati

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

[diff] [blame]

1720

1721

The first version to be created for a model is automatically set as the

1722

default. You must make any subsequent changes to the default version

1723

setting manually using this method.

1724

1725

Args:

1726

name: string, Required. The name of the version to make the default for the model. You

1727

can get the names of all the versions of a model by calling

Dan O'Meara

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

[diff] [blame]

1728

projects.models.versions.list. (required)

Bu Sun Kim

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

[diff] [blame]

1729

body: object, The request body.

Sai Cheemalapati

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

[diff] [blame]

1730

The object takes the form of:

1731

1732

{ # Request message for the SetDefaultVersion request.

1733

}

1734

1735

x__xgafv: string, V1 error format.

Allowed values

1 - v1 error format

2 - v2 error format

Returns:

An object of the form:

1742

1743

{ # Represents a version of the model.

1744

#

1745

# Each version is a trained model deployed in the cloud, ready to handle

1746

# prediction requests. A model can have multiple versions. You can get

1747

# information about all of the versions of a given model by calling

Dan O'Meara

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

[diff] [blame]

1748

# projects.models.versions.list.

Bu Sun Kim

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

[diff] [blame]

1749

"acceleratorConfig": { # Represents a hardware accelerator request config. # Optional. Accelerator config for using GPUs for online prediction (beta).

1750

# Only specify this field if you have specified a Compute Engine (N1) machine

1751

# type in the `machineType` field. Learn more about [using GPUs for online

1752

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

1753

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

1754

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

1755

# [accelerators for online

1756

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

1757

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

1758

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

1759

},

1760

"isDefault": True or False, # Output only. If true, this version will be used to handle prediction

1761

# requests that do not specify a version.

1762

#

1763

# You can change the default version by calling

1764

# projects.methods.versions.setDefault.

Bu Sun Kim

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

[diff] [blame]

1765

"manualScaling": { # Options for manually scaling a model. # Manually select the number of nodes to use for serving the

1766

# model. You should generally use `auto_scaling` with an appropriate

1767

# `min_nodes` instead, but this option is available if you want more

1768

# predictable billing. Beware that latency and error rates will increase

1769

# if the traffic exceeds that capability of the system to serve it based

1770

# on the selected number of nodes.

1771

"nodes": 42, # The number of nodes to allocate for this model. These nodes are always up,

1772

# starting from the time the model is deployed, so the cost of operating

1773

# this model will be proportional to `nodes` * number of hours since

1774

# last billing cycle plus the cost for each prediction performed.

Dan O'Meara

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

[diff] [blame]

1775

},

Bu Sun Kim

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

[diff] [blame]

1776

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

Bu Sun Kim

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

[diff] [blame]

1777

"name": "A String", # Required. The name specified for the version when it was created.

1778

#

1779

# The version name must be unique within the model it is created in.

1780

"serviceAccount": "A String", # Optional. Specifies the service account for resource access control.

1781

"pythonVersion": "A String", # Required. The version of Python used in prediction.

1782

#

1783

# The following Python versions are available:

1784

#

1785

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

1786

# later.

1787

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

1788

# from '1.4' to '1.14'.

1789

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

1790

# earlier.

1791

#

1792

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

1793

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

1794

"lastUseTime": "A String", # Output only. The time the version was last used for prediction.

1795

"predictionClass": "A String", # Optional. The fully qualified name

Dan O'Meara

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

[diff] [blame]

1796

# (<var>module_name</var>.<var>class_name</var>) of a class that implements

Bu Sun Kim

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

[diff] [blame]

1797

# the Predictor interface described in this reference field. The module

1798

# containing this class should be included in a package provided to the

1799

# [`packageUris` field](#Version.FIELDS.package_uris).

Sai Cheemalapati

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

[diff] [blame]

1800

#

Bu Sun Kim

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

[diff] [blame]

1801

# Specify this field if and only if you are deploying a [custom prediction

1802

# routine (beta)](/ml-engine/docs/tensorflow/custom-prediction-routines).

1803

# If you specify this field, you must set

Dan O'Meara

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

[diff] [blame]

1804

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater and

1805

# you must set `machineType` to a [legacy (MLS1)

1806

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

Bu Sun Kim

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

[diff] [blame]

1807

#

1808

# The following code sample provides the Predictor interface:

1809

#

Bu Sun Kim

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

[diff] [blame]

1810

# <pre style="max-width: 626px;">

Bu Sun Kim

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

[diff] [blame]

1811

# class Predictor(object):

Bu Sun Kim

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

[diff] [blame]

1812

# """Interface for constructing custom predictors."""

Bu Sun Kim

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

[diff] [blame]

1813

#

1814

# def predict(self, instances, **kwargs):

Bu Sun Kim

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

[diff] [blame]

1815

# """Performs custom prediction.

Bu Sun Kim

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

[diff] [blame]

1816

#

1817

# Instances are the decoded values from the request. They have already

1818

# been deserialized from JSON.

1819

#

1820

# Args:

1821

# instances: A list of prediction input instances.

1822

# **kwargs: A dictionary of keyword args provided as additional

1823

# fields on the predict request body.

1824

#

1825

# Returns:

1826

# A list of outputs containing the prediction results. This list must

1827

# be JSON serializable.

Bu Sun Kim

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

[diff] [blame]

1828

# """

Bu Sun Kim

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

[diff] [blame]

1829

# raise NotImplementedError()

1830

#

1831

# @classmethod

1832

# def from_path(cls, model_dir):

Bu Sun Kim

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

[diff] [blame]

1833

# """Creates an instance of Predictor using the given path.

Bu Sun Kim

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

[diff] [blame]

1834

#

1835

# Loading of the predictor should be done in this method.

1836

#

1837

# Args:

1838

# model_dir: The local directory that contains the exported model

1839

# file along with any additional files uploaded when creating the

# version resource.

#

# Returns:

# An instance implementing this Predictor class.

Bu Sun Kim

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

[diff] [blame]

1844

# """

Bu Sun Kim

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

[diff] [blame]

1845

# raise NotImplementedError()

Dan O'Meara

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

[diff] [blame]

1846

# </pre>

Bu Sun Kim

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

[diff] [blame]

1847

#

1848

# Learn more about [the Predictor interface and custom prediction

1849

# routines](/ml-engine/docs/tensorflow/custom-prediction-routines).

Bu Sun Kim

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

[diff] [blame]

1850

"deploymentUri": "A String", # Required. The Cloud Storage location of the trained model used to

1851

# create the version. See the

1852

# [guide to model

1853

# deployment](/ml-engine/docs/tensorflow/deploying-models) for more

1854

# information.

1855

#

1856

# When passing Version to

1857

# projects.models.versions.create

1858

# the model service uses the specified location as the source of the model.

1859

# Once deployed, the model version is hosted by the prediction service, so

1860

# this location is useful only as a historical record.

1861

# The total number of model files can't exceed 1000.

Bu Sun Kim

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

[diff] [blame]

1862

"packageUris": [ # Optional. Cloud Storage paths (`gs://…`) of packages for [custom

Bu Sun Kim

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

[diff] [blame]

1863

# prediction routines](/ml-engine/docs/tensorflow/custom-prediction-routines)

1864

# or [scikit-learn pipelines with custom

1865

# code](/ml-engine/docs/scikit/exporting-for-prediction#custom-pipeline-code).

1866

#

1867

# For a custom prediction routine, one of these packages must contain your

1868

# Predictor class (see

1869

# [`predictionClass`](#Version.FIELDS.prediction_class)). Additionally,

1870

# include any dependencies used by your Predictor or scikit-learn pipeline

1871

# uses that are not already included in your selected [runtime

1872

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

1873

#

1874

# If you specify this field, you must also set

1875

# [`runtimeVersion`](#Version.FIELDS.runtime_version) to 1.4 or greater.

Bu Sun Kim

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

[diff] [blame]

1876

"A String",

Bu Sun Kim

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

[diff] [blame]

1877

],

Bu Sun Kim

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

[diff] [blame]

1878

"explanationConfig": { # Message holding configuration options for explaining model predictions. # Optional. Configures explainability features on the model's version.

1879

# Some explanation features require additional metadata to be loaded

1880

# as part of the model payload.

1881

# There are two feature attribution methods supported for TensorFlow models:

1882

# integrated gradients and sampled Shapley.

1883

# [Learn more about feature

1884

# attributions.](/ai-platform/prediction/docs/ai-explanations/overview)

1885

"integratedGradientsAttribution": { # Attributes credit by computing the Aumann-Shapley value taking advantage # Attributes credit by computing the Aumann-Shapley value taking advantage

1886

# of the model's fully differentiable structure. Refer to this paper for

1887

# more details: http://proceedings.mlr.press/v70/sundararajan17a.html

1888

# of the model's fully differentiable structure. Refer to this paper for

1889

# more details: https://arxiv.org/abs/1703.01365

1890

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

1891

# A good value to start is 50 and gradually increase until the

1892

# sum to diff property is met within the desired error range.

1893

},

1894

"sampledShapleyAttribution": { # An attribution method that approximates Shapley values for features that # An attribution method that approximates Shapley values for features that

1895

# contribute to the label being predicted. A sampling strategy is used to

1896

# approximate the value rather than considering all subsets of features.

1897

# contribute to the label being predicted. A sampling strategy is used to

1898

# approximate the value rather than considering all subsets of features.

1899

"numPaths": 42, # The number of feature permutations to consider when approximating the

1900

# Shapley values.

1901

},

1902

"xraiAttribution": { # Attributes credit by computing the XRAI taking advantage # Attributes credit by computing the XRAI taking advantage

1903

# of the model's fully differentiable structure. Refer to this paper for

1904

# more details: https://arxiv.org/abs/1906.02825

1905

# Currently only implemented for models with natural image inputs.

1906

# of the model's fully differentiable structure. Refer to this paper for

1907

# more details: https://arxiv.org/abs/1906.02825

1908

# Currently only implemented for models with natural image inputs.

1909

"numIntegralSteps": 42, # Number of steps for approximating the path integral.

1910

# A good value to start is 50 and gradually increase until the

1911

# sum to diff property is met within the desired error range.

1912

},

1913

},

Bu Sun Kim

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

[diff] [blame]

1914

"autoScaling": { # Options for automatically scaling a model. # Automatically scale the number of nodes used to serve the model in

Dan O'Meara

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

[diff] [blame]

1915

# response to increases and decreases in traffic. Care should be

Bu Sun Kim

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

[diff] [blame]

1916

# taken to ramp up traffic according to the model's ability to scale

Dan O'Meara

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

[diff] [blame]

1917

# or you will start seeing increases in latency and 429 response codes.

1918

#

1919

# Note that you cannot use AutoScaling if your version uses

1920

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use specify

1921

# `manual_scaling`.

Bu Sun Kim

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

[diff] [blame]

1922

"minNodes": 42, # Optional. The minimum number of nodes to allocate for this model. These

Dan O'Meara

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

[diff] [blame]

1923

# nodes are always up, starting from the time the model is deployed.

1924

# Therefore, the cost of operating this model will be at least

1925

# `rate` * `min_nodes` * number of hours since last billing cycle,

1926

# where `rate` is the cost per node-hour as documented in the

1927

# [pricing guide](/ml-engine/docs/pricing),

1928

# even if no predictions are performed. There is additional cost for each

1929

# prediction performed.

1930

#

1931

# Unlike manual scaling, if the load gets too heavy for the nodes

1932

# that are up, the service will automatically add nodes to handle the

1933

# increased load as well as scale back as traffic drops, always maintaining

1934

# at least `min_nodes`. You will be charged for the time in which additional

1935

# nodes are used.

1936

#

1937

# If `min_nodes` is not specified and AutoScaling is used with a [legacy

1938

# (MLS1) machine type](/ml-engine/docs/machine-types-online-prediction),

1939

# `min_nodes` defaults to 0, in which case, when traffic to a model stops

1940

# (and after a cool-down period), nodes will be shut down and no charges will

1941

# be incurred until traffic to the model resumes.

1942

#

1943

# If `min_nodes` is not specified and AutoScaling is used with a [Compute

1944

# Engine (N1) machine type](/ml-engine/docs/machine-types-online-prediction),

1945

# `min_nodes` defaults to 1. `min_nodes` must be at least 1 for use with a

1946

# Compute Engine machine type.

1947

#

1948

# Note that you cannot use AutoScaling if your version uses

1949

# [GPUs](#Version.FIELDS.accelerator_config). Instead, you must use

1950

# ManualScaling.

1951

#

1952

# You can set `min_nodes` when creating the model version, and you can also

1953

# update `min_nodes` for an existing version:

1954

# <pre>

1955

# update_body.json:

1956

# {

Bu Sun Kim

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

[diff] [blame]

1957

# 'autoScaling': {

1958

# 'minNodes': 5

Dan O'Meara

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

[diff] [blame]

# }

# }

# </pre>

# HTTP request:

Bu Sun Kim

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

[diff] [blame]

1963

# <pre style="max-width: 626px;">

Dan O'Meara

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

[diff] [blame]

1964

# PATCH

1965

# https://ml.googleapis.com/v1/{name=projects/*/models/*/versions/*}?update_mask=autoScaling.minNodes

1966

# -d @./update_body.json

1967

# </pre>

1968

},

Bu Sun Kim

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

[diff] [blame]

1969

"createTime": "A String", # Output only. The time the version was created.

Bu Sun Kim

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

[diff] [blame]

1970

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

1971

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

1972

# are arbitrary strings that you supply.

1973

# For more information, see the documentation on

1974

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

1975

"a_key": "A String",

1976

},

Bu Sun Kim

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

[diff] [blame]

1977

"requestLoggingConfig": { # Configuration for logging request-response pairs to a BigQuery table. # Optional. *Only* specify this field in a

Dan O'Meara

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

[diff] [blame]

1978

# projects.models.versions.patch

1979

# request. Specifying it in a

1980

# projects.models.versions.create

1981

# request has no effect.

1982

#

1983

# Configures the request-response pair logging on predictions from this

1984

# Version.

1985

# Online prediction requests to a model version and the responses to these

1986

# requests are converted to raw strings and saved to the specified BigQuery

1987

# table. Logging is constrained by [BigQuery quotas and

1988

# limits](/bigquery/quotas). If your project exceeds BigQuery quotas or limits,

1989

# AI Platform Prediction does not log request-response pairs, but it continues

1990

# to serve predictions.

1991

#

1992

# If you are using [continuous

1993

# evaluation](/ml-engine/docs/continuous-evaluation/), you do not need to

1994

# specify this configuration manually. Setting up continuous evaluation

1995

# automatically enables logging of request-response pairs.

Bu Sun Kim

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

[diff] [blame]

1996

"bigqueryTableName": "A String", # Required. Fully qualified BigQuery table name in the following format:

1997

# "<var>project_id</var>.<var>dataset_name</var>.<var>table_name</var>"

Dan O'Meara

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

[diff] [blame]

1998

#

Bu Sun Kim

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

[diff] [blame]

1999

# The specified table must already exist, and the "Cloud ML Service Agent"

Dan O'Meara

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

[diff] [blame]

2000

# for your project must have permission to write to it. The table must have

2001

# the following [schema](/bigquery/docs/schemas):

2002

#

2003

# <table>

Bu Sun Kim

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

[diff] [blame]

2004

# <tr><th>Field name</th><th style="display: table-cell">Type</th>

2005

# <th style="display: table-cell">Mode</th></tr>

Dan O'Meara

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

[diff] [blame]

2006

# <tr><td>model</td><td>STRING</td><td>REQUIRED</td></tr>

2007

# <tr><td>model_version</td><td>STRING</td><td>REQUIRED</td></tr>

2008

# <tr><td>time</td><td>TIMESTAMP</td><td>REQUIRED</td></tr>

2009

# <tr><td>raw_data</td><td>STRING</td><td>REQUIRED</td></tr>

2010

# <tr><td>raw_prediction</td><td>STRING</td><td>NULLABLE</td></tr>

2011

# <tr><td>groundtruth</td><td>STRING</td><td>NULLABLE</td></tr>

2012

# </table>

Bu Sun Kim

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

[diff] [blame]

2013

"samplingPercentage": 3.14, # Percentage of requests to be logged, expressed as a fraction from 0 to 1.

2014

# For example, if you want to log 10% of requests, enter `0.1`. The sampling

2015

# window is the lifetime of the model version. Defaults to 0.

2016

},

2017

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

2018

"machineType": "A String", # Optional. The type of machine on which to serve the model. Currently only

2019

# applies to online prediction service. If this field is not specified, it

2020

# defaults to `mls1-c1-m2`.

2021

#

2022

# Online prediction supports the following machine types:

#

# * `mls1-c1-m2`

# * `mls1-c4-m2`

# * `n1-standard-2`

# * `n1-standard-4`

# * `n1-standard-8`

# * `n1-standard-16`

# * `n1-standard-32`

# * `n1-highmem-2`

# * `n1-highmem-4`

# * `n1-highmem-8`

# * `n1-highmem-16`

# * `n1-highmem-32`

# * `n1-highcpu-2`

# * `n1-highcpu-4`

# * `n1-highcpu-8`

# * `n1-highcpu-16`

# * `n1-highcpu-32`

#

# `mls1-c1-m2` is generally available. All other machine types are available

2043

# in beta. Learn more about the [differences between machine

2044

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

2045

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

2046

#

2047

# For more information, see the

2048

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

2049

# [how to manage runtime versions](/ml-engine/docs/versioning).

2050

"description": "A String", # Optional. The description specified for the version when it was created.

2051

"framework": "A String", # Optional. The machine learning framework AI Platform uses to train

2052

# this version of the model. Valid values are `TENSORFLOW`, `SCIKIT_LEARN`,

2053

# `XGBOOST`. If you do not specify a framework, AI Platform

2054

# will analyze files in the deployment_uri to determine a framework. If you

2055

# choose `SCIKIT_LEARN` or `XGBOOST`, you must also set the runtime version

2056

# of the model to 1.4 or greater.

2057

#

2058

# Do **not** specify a framework if you're deploying a [custom

2059

# prediction routine](/ml-engine/docs/tensorflow/custom-prediction-routines).

2060

#

2061

# If you specify a [Compute Engine (N1) machine

2062

# type](/ml-engine/docs/machine-types-online-prediction) in the

2063

# `machineType` field, you must specify `TENSORFLOW`

2064

# for the framework.

2065

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

2066

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

2067

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

2068

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

2069

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

2070

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

2071

# ensure that their change will be applied to the model as intended.

Sai Cheemalapati