blob: 503f8318485b0e0d9bd9e2e0710f93b6d9744c33 [file] [log] [blame] [view]
Vijay Vasudevan2d0d1262016-08-08 14:06:20 -08001# Changes since last release
2
3## Breaking Changes to the API
4
Eugene Brevdo21e1cc72016-08-11 21:45:39 -08005* LSTMCell, BasicLSTMCell, and MultiRNNCell constructors now default to
6 `state_is_tuple=True`. For a quick fix while transitioning to the new
7 default, simply pass the argument `state_is_tuple=False`.
Vijay Vasudevan2d0d1262016-08-08 14:06:20 -08008* DeviceFactory's AddDevices and CreateDevices functions now return
9 a Status instead of void.
A. Unique TensorFlower84cefad2016-08-12 07:06:13 -080010* Int32 elements of list(type) arguments are no longer placed in host memory by
11 default. If necessary, a list(type) argument to a kernel can be placed in host
12 memory using a HostMemory annotation.
Wei Ho829a2362016-08-24 15:26:40 -080013* uniform_unit_scaling_initializer() no longer takes a full_shape arg, instead
14 relying on the partition info passed to the initializer function when it's
15 called.
A. Unique TensorFlower672eebd2016-08-25 07:19:38 -080016* The NodeDef protocol message is now defined in its own file node_def.proto
17 instead of graph.proto.
Vijay Vasudevan612bae72016-09-09 11:03:09 -080018* ops.NoGradient was renamed ops.NotDifferentiable. ops.NoGradient will
19 be removed soon.
Dan Manée5bcf542016-05-16 13:39:34 -080020
A. Unique TensorFlowerabe9ab32016-07-31 22:07:30 -080021# Release 0.10.0
A. Unique TensorFlower533d8912016-06-30 12:10:50 -080022
A. Unique TensorFlowerabe9ab32016-07-31 22:07:30 -080023## Major Features and Improvements
24
25* Added support for C++ shape inference
26* Added graph-construction C API
27* Major revision to the graph-construction C++ API
28* Support makefile build for iOS
29* Added Mac GPU support
30* Full version of TF-Slim available as `tf.contrib.slim`
31* Added k-Means clustering and WALS matrix factorization
32
33## Big Fixes and Other Changes
34
35* Allow gradient computation for scalar values.
36* Performance improvements for gRPC
37* Improved support for fp16
38* New high-level ops in tf.contrib.{layers,metrics}
39* New features for TensorBoard, such as shape display, exponential smoothing
40* Faster and more stable Google Cloud Storage (GCS) filesystem support
41* Support for zlib compression and decompression for TFRecordReader and TFRecordWriter
42* Support for reading (animated) GIFs
43* Improved support for SparseTensor
44* Added support for more probability distributions (Dirichlet, Beta, Bernoulli, etc.)
45* Added Python interfaces to reset resource containers.
46* Many bugfixes and performance improvements
47* Many documentation fixes
48
49## Thanks to our Contributors
50
51This release contains contributions from many people at Google, as well as:
52
53Alex Rothberg, Andrew Royer, Austin Marshall, @BlackCoal, Bob Adolf, Brian Diesel, Charles-Emmanuel Dias, @chemelnucfin, Chris Lesniewski, Daeyun Shin, Daniel Rodriguez, Danijar Hafner, Darcy Liu, Kristinn R. Thórisson, Daniel Castro, Dmitry Savintsev, Kashif Rasul, Dylan Paiton, Emmanuel T. Odeke, Ernest Grzybowski, Gavin Sherry, Gideon Dresdner, Gregory King, Harold Cooper, @heinzbeinz, Henry Saputra, Huarong Huo, Huazuo Gao, Igor Babuschkin, Igor Macedo Quintanilha, Ivan Ukhov, James Fysh, Jan Wilken Dörrie, Jihun Choi, Johnny Lim, Jonathan Raiman, Justin Francis, @lilac, Li Yi, Marc Khoury, Marco Marchesi, Max Melnick, Micael Carvalho, @mikowals, Mostafa Gazar, Nico Galoppo, Nishant Agrawal, Petr Janda, Yuncheng Li, @raix852, Robert Rose, @Robin-des-Bois, Rohit Girdhar, Sam Abrahams, satok16, Sergey Kishchenko, Sharkd Tu, @shotat, Siddharth Agrawal, Simon Denel, @sono-bfio, SunYeop Lee, Thijs Vogels, @tobegit3hub, @Undo1, Wang Yang, Wenjian Huang, Yaroslav Bulatov, Yuan Tang, Yunfeng Wang, Ziming Dong
54
55We are also grateful to all who filed issues or helped resolve them, asked and
56answered questions, and were part of inspiring discussions.
Vijay Vasudevan490afa92016-06-21 09:18:06 -080057
58# Release 0.9.0
59
60## Major Features and Improvements
61
62* Python 3.5 support and binaries
63* Added iOS support
64* Added support for processing on GPUs on MacOS
65* Added makefile for better cross-platform build support (C API only)
66* fp16 support and improved complex128 support for many ops
67* Higher level functionality in contrib.{layers,losses,metrics,learn}
68* More features to Tensorboard
69* Improved support for string embedding and sparse features
70* The RNN api is finally "official" (see, e.g., `tf.nn.dynamic_rnn`,
71 `tf.nn.rnn`, and the classes in `tf.nn.rnn_cell`).
72* TensorBoard now has an Audio Dashboard, with associated audio summaries.
73
74## Big Fixes and Other Changes
75
76* Turned on CuDNN Autotune.
77* Added support for using third-party Python optimization algorithms (contrib.opt).
78* Google Cloud Storage filesystem support.
79* HDF5 support
80* Add support for 3d convolutions and pooling.
81* Update gRPC release to 0.14.
82* Eigen version upgrade.
83* Switch to eigen thread pool
84* `tf.nn.moments()` now accepts a `shift` argument. Shifting by a good estimate
85 of the mean improves numerical stability. Also changes the behavior of the
86 `shift` argument to `tf.nn.sufficient_statistics()`.
87* Performance improvements
88* Many bugfixes
89* Many documentation fixes
90* TensorBoard fixes: graphs with only one data point, Nan values,
91 reload button and auto-reload, tooltips in scalar charts, run
92 filtering, stable colors
93* Tensorboard graph visualizer now supports run metadata. Clicking on nodes
94 while viewing a stats for a particular run will show runtime statistics, such
95 as memory or compute usage. Unused nodes will be faded out.
96
97## Thanks to our Contributors
98
99This release contains contributions from many people at Google, as well as:
100
A. Unique TensorFlowerabe9ab32016-07-31 22:07:30 -0800101Aaron Schumacher, Aidan Dang, Akihiko ITOH, Aki Sukegawa, Arbit Chen, Aziz Alto, Danijar Hafner, Erik Erwitt, Fabrizio Milo, Felix Maximilian Möller, Henry Saputra, Sung Kim, Igor Babuschkin, Jan Zikes, Jeremy Barnes, Jesper Steen Møller, Johannes Mayer, Justin Harris, Kashif Rasul, Kevin Robinson, Loo Rong Jie, Lucas Moura, Łukasz Bieniasz-Krzywiec, Mario Cho, Maxim Grechkin, Michael Heilman, Mostafa Rahmani, Mourad Mourafiq, @ninotoshi, Orion Reblitz-Richardson, Yuncheng Li, @raoqiyu, Robert DiPietro, Sam Abrahams, Sebastian Raschka, Siddharth Agrawal, @snakecharmer1024, Stephen Roller, Sung Kim, SunYeop Lee, Thijs Vogels, Till Hoffmann, Victor Melo, Ville Kallioniemi, Waleed Abdulla, Wenjian Huang, Yaroslav Bulatov, Yeison Rodriguez, Yuan Tang, Yuxin Wu, @zhongzyd, Ziming Dong, Zohar Jackson
Vijay Vasudevan490afa92016-06-21 09:18:06 -0800102
103We are also grateful to all who filed issues or helped resolve them, asked and
104answered questions, and were part of inspiring discussions.
105
Illia Polosukhin5c9bc512016-04-18 17:56:51 -0800106# Release 0.8.0
107
108## Major Features and Improvements
109
110* Added a distributed runtime using GRPC
111* Move skflow to `contrib/learn`
112* Better linear optimizer in `contrib/linear_optimizer`
113* Random forest implementation in `contrib/tensor_forest`
114* CTC loss and decoders in `contrib/ctc`
115* Basic support for `half` data type
116* Better support for loading user ops (see examples in `contrib/`)
117* Allow use of (non-blocking) Eigen threadpool with `TENSORFLOW_USE_EIGEN_THREADPOOL` define
118* Add an extension mechanism for adding network file system support
119* TensorBoard displays metadata stats (running time, memory usage and device used) and tensor shapes
120
121## Big Fixes and Other Changes
122
123* Utility for inspecting checkpoints
124* Basic tracing and timeline support
125* Allow building against cuDNN 5 (not incl. RNN/LSTM support)
126* Added instructions and binaries for ProtoBuf library with fast serialization and without 64MB limit
127* Added special functions
128* `bool`-strictness: Tensors have to be explictly compared to `None`
129* Shape strictness: all fed values must have a shape that is compatible with the tensor they are replacing
130* Exposed `tf.while_loop` (deprecated `control_flow_ops.While`)
131* run() now takes RunOptions and RunMetadata, which enable timing stats
132* Fixed lots of potential overflow problems in op kernels
133* Various performance improvements, especially for RNNs and convolutions
134* Many bugfixes
135* Nightly builds, tutorial tests, many test improvements
136* New examples: transfer learning and deepdream ipython notebook
137* Added tutorials, many documentation fixes.
138
139## Thanks to our Contributors
140
141This release contains contributions from many people at Google, as well as:
142
A. Unique TensorFlowerabe9ab32016-07-31 22:07:30 -0800143Abhinav Upadhyay, Aggelos Avgerinos, Alan Wu, Alexander G. de G. Matthews, Aleksandr Yahnev, @amchercashin, Andy Kitchen, Aurelien Geron, Awni Hannun, @BanditCat, Bas Veeling, Cameron Chen, @cg31, Cheng-Lung Sung, Christopher Bonnett, Dan Becker, Dan Van Boxel, Daniel Golden, Danijar Hafner, Danny Goodman, Dave Decker, David Dao, David Kretch, Dongjoon Hyun, Dustin Dorroh, @e-lin, Eurico Doirado, Erik Erwitt, Fabrizio Milo, @gaohuazuo, Iblis Lin, Igor Babuschkin, Isaac Hodes, Isaac Turner, Iván Vallés, J Yegerlehner, Jack Zhang, James Wexler, Jan Zikes, Jay Young, Jeff Hodges, @jmtatsch, Johnny Lim, Jonas Meinertz Hansen, Kanit Wongsuphasawat, Kashif Rasul, Ken Shirriff, Kenneth Mitchner, Kenta Yonekura, Konrad Magnusson, Konstantin Lopuhin, @lahwran, @lekaha, @liyongsea, Lucas Adams, @makseq, Mandeep Singh, @manipopopo, Mark Amery, Memo Akten, Michael Heilman, Michael Peteuil, Nathan Daly, Nicolas Fauchereau, @ninotoshi, Olav Nymoen, @panmari, @papelita1234, Pedro Lopes, Pranav Sailesh Mani, RJ Ryan, Rob Culliton, Robert DiPietro, @ronrest, Sam Abrahams, Sarath Shekkizhar, Scott Graham, Sebastian Raschka, Sung Kim, Surya Bhupatiraju, Syed Ahmed, Till Hoffmann, @timsl, @urimend, @vesnica, Vlad Frolov, Vlad Zagorodniy, Wei-Ting Kuo, Wenjian Huang, William Dmitri Breaden Madden, Wladimir Schmidt, Yuan Tang, Yuwen Yan, Yuxin Wu, Yuya Kusakabe, @zhongzyd, @znah.
Illia Polosukhin5c9bc512016-04-18 17:56:51 -0800144
145We are also grateful to all who filed issues or helped resolve them, asked and
146answered questions, and were part of inspiring discussions.
147
148
Eugene Brevdo56f1d642016-03-10 17:18:30 -0800149# Release 0.7.1
150
151## Bug Fixes and Other Changes
152
153* Added gfile.Open and gfile.Copy, used by input_data.py.
154* Fixed Saver bug when MakeDirs tried to create empty directory.
155* GPU Pip wheels are built with cuda 7.5 and cudnn-v4, making them
156 required for the binary releases. Lower versions of cuda/cudnn can
157 be supported by installing from sources and setting the options
158 during ./configure
159* Fix dataset encoding example for Python3 (@danijar)
160* Fix PIP installation by not packaging protobuf as part of wheel,
161 require protobuf 3.0.0b2.
162* Fix Mac pip installation of numpy by requiring pip >= 1.10.1.
163* Improvements and fixes to Docker image.
164
165
Vijay Vasudevanfe056f02016-02-17 11:42:30 -0800166# Release 0.7.0
Vijay Vasudevan10e62dc2015-12-11 23:03:16 -0800167
Vijay Vasudevanfe056f02016-02-17 11:42:30 -0800168## Major Features and Improvements
169
170* Allow using any installed Cuda >= 7.0 and cuDNN >= R2, and add support
171 for cuDNN R4
172* Added a `contrib/` directory for unsupported or experimental features,
173 including higher level `layers` module
174* Added an easy way to add and dynamically load user-defined ops
175* Built out a good suite of tests, things should break less!
176* Added `MetaGraphDef` which makes it easier to save graphs with metadata
177* Added assignments for "Deep Learning with TensorFlow" udacity course
178
179
180## Bug Fixes and Other Changes
181
182* Added a versioning framework for `GraphDef`s to ensure compatibility
183* Enforced Python 3 compatibility
184* Internal changes now show up as sensibly separated commits
185* Open-sourced the doc generator
186* Un-fork Eigen
187* Simplified the `BUILD` files and cleaned up C++ headers
188* TensorFlow can now be used as a submodule in another bazel build
189* New ops (e.g., `*fft`, `*_matrix_solve`)
190* Support for more data types in many ops
191* Performance improvements
192* Various bugfixes
193* Documentation fixes and improvements
194
195
196## Breaking Changes to the API
Vijay Vasudevan10e62dc2015-12-11 23:03:16 -0800197
198* `AdjustContrast` kernel deprecated, new kernel `AdjustContrastv2` takes and
199 outputs float only. `adjust_contrast` now takes all data types.
200* `adjust_brightness`'s `delta` argument is now always assumed to be in `[0,1]`
201 (as is the norm for images in floating point formats), independent of the
202 data type of the input image.
203* The image processing ops do not take `min` and `max` inputs any more, casting
204 safety is handled by `saturate_cast`, which makes sure over- and underflows
205 are handled before casting to data types with smaller ranges.
Geoffrey Irvingcbff45c2016-01-12 08:06:56 -0800206* For C++ API users: `IsLegacyScalar` and `IsLegacyVector` are now gone from
207 `TensorShapeUtils` since TensorFlow is scalar strict within Google (for
208 example, the shape argument to `tf.reshape` can't be a scalar anymore). The
209 open source release was already scalar strict, so outside Google `IsScalar`
210 and `IsVector` are exact replacements.
Josh Levenbergdb7478e2016-01-20 14:54:50 -0800211* The following files are being removed from `tensorflow/core/public/`:
212 * `env.h` -> `../platform/env.h`
213 * `status.h` -> `../lib/core/status.h`
214 * `tensor.h` -> `../framework/tensor.h`
215 * `tensor_shape.h` -> `../framework/tensor_shape.h`
216 * `partial_tensor_shape.h` -> `../framework/partial_tensor_shape.h`
217 * `tensorflow_server.h` deleted
Geoffrey Irving56437752016-01-25 09:43:13 -0800218* For C++ API users: `TensorShape::ShortDebugString` has been renamed to
219 `DebugString`, and the previous `DebugString` behavior is gone (it was
220 needlessly verbose and produced a confusing empty string for scalars).
Manjunath Kudlurc2722a12016-01-27 13:24:50 -0800221* `GraphOptions.skip_common_subexpression_elimination` has been removed. All
222 graph optimizer options are now specified via
223 `GraphOptions.OptimizerOptions`.
Geoffrey Irving18297122016-02-10 11:48:34 -0800224* `ASSERT_OK` / `EXPECT_OK` macros conflicted with external projects, so they
225 were renamed `TF_ASSERT_OK`, `TF_EXPECT_OK`. The existing macros are
226 currently maintained for short-term compatibility but will be removed.
Eugene Brevdofea55e12016-01-27 14:54:54 -0800227* The non-public `nn.rnn` and the various `nn.seq2seq` methods now return
228 just the final state instead of the list of all states.
Vijay Vasudevanfe056f02016-02-17 11:42:30 -0800229* `tf.scatter_update` now no longer guarantees that lexicographically largest
230 index be used for update when duplicate entries exist.
Geoffrey Irving3e33d442016-02-08 12:02:44 -0800231* `tf.image.random_crop(image, [height, width])` is now
232 `tf.random_crop(image, [height, width, depth])`, and `tf.random_crop` works
233 for any rank (not just 3-D images). The C++ `RandomCrop` op has been replaced
234 with pure Python.
Geoffrey Irving18297122016-02-10 11:48:34 -0800235* Renamed `tf.test.GetTempDir` and `tf.test.IsBuiltWithCuda` to
236 `tf.test.get_temp_dir` and `tf.test.is_built_with_cuda` for PEP-8
237 compatibility.
Vijay Vasudevanfe056f02016-02-17 11:42:30 -0800238* `parse_example`'s interface has changed, the old interface is accessible in
239 `legacy_parse_example` (same for related functions).
240* New `Variable`s are not added to the same collection several times even if
241 a list with duplicates is passed to the constructor.
Josh Levenberg02dff6d2016-01-07 18:37:54 -0800242* The Python API will now properly set the `list` member of `AttrValue` in
243 constructed `GraphDef` messages for empty lists. The serialization of some
244 graphs will change, but the change is both forwards and backwards compatible.
245 It will break tests that compare a generated `GraphDef` to a golden serialized
Vijay Vasudevanfe056f02016-02-17 11:42:30 -0800246 `GraphDef` (which is discouraged).
247
248
249## Thanks to our Contributors
250
251This release contains contributions from many people at Google, as well as:
252
253Akiomi Kamakura, Alex Vig, Alexander Rosenberg Johansen, Andre Cruz, Arun Ahuja,
254Bart Coppens, Bernardo Pires, Carl Vondrick, Cesar Salgado, Chen Yu,
255Christian Jauvin, Damien Aymeric, Dan Vanderkam, Denny Britz, Dongjoon Hyun,
256Eren Güven, Erik Erwitt, Fabrizio Milo, G. Hussain Chinoy, Jim Fleming,
257Joao Felipe Santos, Jonas Meinertz Hansen, Joshi Rekha, Julian Viereck,
258Keiji Ariyama, Kenton Lee, Krishna Sankar, Kristina Chodorow, Linchao Zhu,
259Lukas Krecan, Mark Borgerding, Mark Daoust, Moussa Taifi,
260Nathan Howell, Naveen Sundar Govindarajulu, Nick Sweeting, Niklas Riekenbrauck,
261Olivier Grisel, Patrick Christ, Povilas Liubauskas, Rainer Wasserfuhr,
262Romain Thouvenin, Sagan Bolliger, Sam Abrahams, Taehoon Kim, Timothy J Laurent,
263Vlad Zavidovych, Yangqing Jia, Yi-Lin Juang, Yuxin Wu, Zachary Lipton,
264Zero Chen, Alan Wu, @brchiu, @emmjaykay, @jalammar, @Mandar-Shinde,
265@nsipplswezey, @ninotoshi, @panmari, @prolearner and @rizzomichaelg.
266
267We are also grateful to all who filed issues or helped resolve them, asked and
268answered questions, and were part of inspiring discussions.
Josh Levenberg02dff6d2016-01-07 18:37:54 -0800269
Geoffrey Irvingcbff45c2016-01-12 08:06:56 -0800270
Vijay Vasudevan2c3738d2015-12-08 14:55:13 -0800271# Release 0.6.0
272
273## Major Features and Improvements
274
275* Python 3.3+ support via changes to python codebase and ability
276 to specify python version via ./configure.
277
278* Some improvements to GPU performance and memory usage:
279 [convnet benchmarks](https://github.com/soumith/convnet-benchmarks/issues/66)
280 roughly equivalent with native cudnn v2 performance. Improvements mostly due
281 to moving to 32-bit indices, faster shuffling kernels. More improvements to
282 come in later releases.
283
284
Vijay Vasudevanfe056f02016-02-17 11:42:30 -0800285## Bug Fixes
Vijay Vasudevan2c3738d2015-12-08 14:55:13 -0800286
287* Lots of fixes to documentation and tutorials, many contributed
288 by the public.
289
290* 271 closed issues on github issues.
291
Vijay Vasudevanfe056f02016-02-17 11:42:30 -0800292## Backwards-Incompatible Changes
Vijay Vasudevan2c3738d2015-12-08 14:55:13 -0800293
Geoffrey Irving18297122016-02-10 11:48:34 -0800294* `tf.nn.fixed_unigram_candidate_sampler` changed its default 'distortion'
Vijay Vasudevan2c3738d2015-12-08 14:55:13 -0800295 attribute from 0.0 to 1.0. This was a bug in the original release
296 that is now fixed.
297
Vijay Vasudevanddd4aaf2015-12-08 09:58:59 -0800298# Release 0.5.0
299
300Initial release of TensorFlow.