New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Custom configuration of tf.estimator.Estimator - unfavorable change on tf1.1 #9795
Comments
@martinwicke Can you comment? Is this a regression? |
Yes. It's a temporary lockdown. An already existing change will fix this by providing a .replace() method to allow mutations. @xiejw FYI |
It is fixed at head. You can do new_config = run_config.replace( |
(note it may take a day or two to hit github) |
ah, never mind -- but that only works if you use |
commit 39d2e7b17973a820b489292fcceb88959122ee7e Merge: 11a174e a66a1fb Author: Jeroen Bédorf <jeroen@minds.ai> Date: Fri May 12 09:20:42 2017 -0700 Merge branch 'master' of https://github.com/tensorflow/tensorflow into MPIV4 commit a66a1fbeeb779a5366e059a1f97554bc5260ee14 Merge: 85f8a74 2526394 Author: Benoit Steiner <benoitsteiner@users.noreply.github.com> Date: Fri May 12 08:06:48 2017 -0700 Merge pull request #9830 from freedomtan/no-logtostderror There is no --logtostderror commit 85f8a74b2fc288b7c468960dd03fd2a3f604ff85 Merge: 1ffc07e 31d3881 Author: Benoit Steiner <benoitsteiner@users.noreply.github.com> Date: Fri May 12 08:06:01 2017 -0700 Merge pull request #9850 from benoitsteiner/branch_155393864 Branch 155393864 commit 1ffc07e3dcab201e6cff239770108995f03da1d1 Author: Xiaolin Lin <xlin615@gmail.com> Date: Fri May 12 07:03:45 2017 -0700 correct the article (a->an) used before ubuntu (uu-boon-tuu) (#9769) * correct the article used before ubuntu (uu-boon-tuu) * note for adding additional options in compiling CUDA kernel commit 31d3881b4b276b5aab004e69129f05d3a4c51174 Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Fri May 12 06:53:00 2017 -0700 Fixed build error commit bae71921dfc6efb9880add651aaee1292715a282 Author: Paul Van Eck <pvaneck@us.ibm.com> Date: Fri May 12 06:35:31 2017 -0700 Fix how-to reference in distributed runtime README (#9772) The current link 404s, and now that the website documentation has been updated, the link can now point there. Signed-off-by: Paul Van Eck <pvaneck@us.ibm.com> commit 613a0150e5daecfa22faa4d728b6c0de598a41d7 Merge: a33022c 183e2c9 Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Thu May 11 22:46:25 2017 -0700 Merge commit for internal changes commit 183e2c953a81e1a5a7a8b9df2534778b5ae48379 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 22:33:20 2017 -0700 [XLA] Emit FCMP_UNE (true if unordered) instead of FCMP_ONE (false if unordered) for not_equal operation to be more compliant with TF and other languages. PiperOrigin-RevId: 155835642 commit a33022c1470ce1334766b0cad38d9e91c17a2e5d Author: Liangliang He <lliang.he@gmail.com> Date: Fri May 12 13:11:33 2017 +0800 Fix verbs compile error (#9791) commit eb0a59f003977332b7363526fcfd0b3f4311492b Author: Yuan (Terry) Tang <terrytangyuan@users.noreply.github.com> Date: Fri May 12 01:10:39 2017 -0400 Fixes #9654: Allow model_fn being a member function of a class (#9807) * Fixes #9654: Allow model_fn being a member function of a class * Remove self from arg instead of changing a const commit e7f64e391ad4c597392f87711560215186fae2d4 Merge: 33cd434 7d8ea3d Author: Benoit Steiner <benoitsteiner@users.noreply.github.com> Date: Thu May 11 22:09:34 2017 -0700 Merge pull request #9825 from kemaswill/FixBrokenLink_20170510 Fix a broken link commit 33cd434e24e88ef33b74f7d3bd9f7d34034ee05e Merge: dbb78c7 51caa2b Author: Benoit Steiner <benoitsteiner@users.noreply.github.com> Date: Thu May 11 22:08:17 2017 -0700 Merge pull request #9844 from chris-chris/patch/170511-misspell Fix misspells on c++ codes. commit 688f5a6c77bd97e116f55e130872ea0713f9cdb1 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 21:32:32 2017 -0700 [XLA] Add test cases for implicit broadcasting. PiperOrigin-RevId: 155832642 commit 7d8ea3d8c7577e9a674892c57327f8084a87e6e1 Author: Junwei Pan <pandevirus@gmail.com> Date: Wed May 10 20:09:31 2017 -0700 Fix broken links commit d0afbc17dd4afc859bbcdf014d76d0756ba1cd3d Author: Justine Tunney <jart@google.com> Date: Thu May 11 19:21:51 2017 -0700 Use new mirror.bazel.build hostname PiperOrigin-RevId: 155826248 commit cfdf449ae2daa749e66565f846808bb9e5c1bff5 Author: David Soergel <soergel@google.com> Date: Thu May 11 19:19:46 2017 -0700 Update TensorForest to provide output_alternatives for serving. Also, fix reversed inference dict key constants. PiperOrigin-RevId: 155826149 commit 8364dd7d8c9a1ff1ea311b3e70c120ca34215f94 Author: Wei Ho <weiho@google.com> Date: Thu May 11 19:08:52 2017 -0700 Adds warning about non-determinism in returned TensorSlice ordering. PiperOrigin-RevId: 155825601 commit dbb78c7783c32f8f3dc86364dc10d7d51cf61277 Author: Mortada Mehyar <mortada@users.noreply.github.com> Date: Thu May 11 18:55:02 2017 -0700 Fix typos in high performance models docs (#9845) commit 3f575b42994949f0d52b71cac84570b74f2d13c6 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 18:18:23 2017 -0700 Migrate the tf_graph_app component to use webfiles. This is the last of the graph-related components to migrate. This component is not used by TensorBoard, but rather by other apps (such as colab) to embed the graph explorer. PiperOrigin-RevId: 155822603 commit a5b9ef0ec0e9a4f5c42157755ffd180a90063e2b Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 18:17:37 2017 -0700 Add timeline support in tfprof This CL mainly adds timeline support in three views of tfprof. It includes a few other small changes: 1. Handle the case that one Op fires multiple kernels. 2. Remove the requirements for CostGraph for easier user adoption, for now. 3. Some speed improvements in graph view. 4. Consolidate the all kinds of tfprof output into one -output option. PiperOrigin-RevId: 155822542 commit 14221c1fd4155d80341feea61896a4d8abc7ae78 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 18:10:02 2017 -0700 Determine the version of sed based on the output of "--version" rather than ${OSTYPE}. The later approach is not reliable when an user installed GNU sed in Darwin systems like macOS. PiperOrigin-RevId: 155822030 commit 5f9facb82605d4d98e1ace503f683b96f1468446 Author: Justine Tunney <jart@google.com> Date: Thu May 11 17:41:58 2017 -0700 Add Polymer iron-component-page dependency This is a dependency of tf_graph in Tensorboard. PiperOrigin-RevId: 155819863 commit f53c5dad3f37fa9e1503dccadb8714a089f3df87 Merge: 554b57f adb226e Author: Benoit Steiner <benoitsteiner@users.noreply.github.com> Date: Thu May 11 17:27:18 2017 -0700 Merge pull request #9848 from benoitsteiner/branch_155393864 Branch 155393864 commit cb56822129b71afd6009e34e074f90cbcc5bea25 Author: Andrew Harp <andrewharp@google.com> Date: Thu May 11 17:15:26 2017 -0700 Clarify in java/README.md that the instructions there do not apply to Android. PiperOrigin-RevId: 155817593 commit 6487cc13037a06c5223bdd4c1457fe88277c4dca Author: Justine Tunney <jart@google.com> Date: Thu May 11 17:10:36 2017 -0700 Remove all-imports.html reference It's better to directly reference what we need, rather than going through proxy import files. PiperOrigin-RevId: 155817083 commit 1bc8c7e757e9d3af3c9ca4fb5abc8d4bd55922fb Author: Shanqing Cai <cais@google.com> Date: Thu May 11 17:02:33 2017 -0700 tfdbg: reduce the amount of computation in examples_test.sh PiperOrigin-RevId: 155816174 commit 6f5f9838947e7f00681067f321eed7802e8eae87 Author: Eli Bendersky <eliben@google.com> Date: Thu May 11 16:44:46 2017 -0700 [XLA] Improve the documentation of our service/client classes a bit Also kill dead code PiperOrigin-RevId: 155814264 commit adb226ed50cd6a9776970a513a72085e95d7b095 Merge: 554b57f 98ccdcd Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Thu May 11 16:32:55 2017 -0700 Merge commit for internal changes commit 98ccdcd2a0a0835a2804a00d66fcd6014df099bc Author: Benoit Steiner <bsteiner@google.com> Date: Thu May 11 16:23:15 2017 -0700 Automated g4 rollback of changelist 155779520 PiperOrigin-RevId: 155811666 commit 7c2a524feb9eb7623fa5f60107593fc7347a03e9 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 16:13:37 2017 -0700 Android Example: Adding CMake for Gradle build PiperOrigin-RevId: 155810524 commit 90b32e54da641f023e72fc80222bf2acbfb57a4c Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 16:02:56 2017 -0700 Rename some fields, from discussion in go/tf-feature-column and other cls. PiperOrigin-RevId: 155809199 commit a9c50eabd8f19d9695a5b017c7a33921fbbd8f78 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 15:44:08 2017 -0700 Fixes markdown formatting. PiperOrigin-RevId: 155806609 commit 33d977e9c95b57dd0690d716432742a3ab71af4c Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 15:40:04 2017 -0700 Making tf.case deterministic when given an unordered dictionary of predicates. PiperOrigin-RevId: 155805373 commit e52fc916bc23e0fe99ccc5f2cf76bccfabfbd932 Author: Benoit Steiner <bsteiner@google.com> Date: Thu May 11 15:09:48 2017 -0700 Improved regression test PiperOrigin-RevId: 155801810 commit 61d56290b0172e8a20aaad9a1d1c7f7f78fa5b6a Author: Justine Tunney <jart@google.com> Date: Thu May 11 15:06:47 2017 -0700 Use coarse grained namespace for storage PiperOrigin-RevId: 155801412 commit 4e3e8ca367eb9203ba2df07a3826be0005c18157 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 14:48:21 2017 -0700 Handle 1d weights. PiperOrigin-RevId: 155799189 commit dce5a71d588079c86f74033a06e600fb7710ab9b Author: Mustafa Ispir <ispir@google.com> Date: Thu May 11 14:09:33 2017 -0700 FeatureColumn: Improved error message if there is a batch_size mismatch and shapes known at compile time. PiperOrigin-RevId: 155794598 commit bf0363bf3f5a32993f7275c2f5a55ec2962da4a3 Author: Ian Langmore <langmore@google.com> Date: Thu May 11 14:08:31 2017 -0700 MultivariateNormalFullCovariance added to contrib/distributions/ This MVN is parameterized by the full covariance matrix, *not* the square root (as the others are). PiperOrigin-RevId: 155794471 commit 554b57f74a99ea18baf616ae5e3cff8b137430e6 Merge: eb11d6b e838f33 Author: Yifei Feng <fengyifei2026@gmail.com> Date: Thu May 11 14:01:04 2017 -0700 Merge pull request #9843 from benoitsteiner/branch_155393864 Branch 155393864 commit 20f9b5da8d7477c6cc86ee24fd1c78eb41f789f0 Author: Sanders Kleinfeld <skleinfeld@google.com> Date: Thu May 11 13:52:58 2017 -0700 Update to Custom Estimators docs to use input_fn for both .fit() and .evaluate(). Follow-up to PR #9617 PiperOrigin-RevId: 155792276 commit 20be1353372dbb6f2db4b152a37453a2d4e0af14 Author: Anna R <annarev@google.com> Date: Thu May 11 13:45:36 2017 -0700 Added sample benchmark. PiperOrigin-RevId: 155791315 commit 3c2dc3baaae762b00d90761f47265411f54033b3 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 13:27:35 2017 -0700 Renames _parse_example_config to _parse_example_spec, adds tests and other cleanup. PiperOrigin-RevId: 155788799 commit e1e820efb0e46dafd70d8a776b26962927e64454 Author: Jianwei Xie <xiejw@google.com> Date: Thu May 11 13:15:04 2017 -0700 Improve the docstring for learn_runner with new experiment_fn. PiperOrigin-RevId: 155787311 commit e838f3302e1329c5f6cd9a8004e54aa0a5c1cae3 Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Thu May 11 12:36:20 2017 -0700 Turned a bazel dependency error into a warning to keep the build going commit 6479ba550226263e1da8a58ad6c81095693e2751 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 12:24:44 2017 -0700 Go: Update generated wrapper functions for TensorFlow ops. PiperOrigin-RevId: 155780999 commit 011da3f1c022237a5d68601a2a318cff45f0a38d Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 12:18:25 2017 -0700 Update ops-related pbtxt files. PiperOrigin-RevId: 155780263 commit b37da7a32c366d4163da9ed42ed9ab0c84e695be Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 12:12:13 2017 -0700 Add timeline support in tfprof This CL mainly adds timeline support in three views of tfprof. It includes a few other small changes: 1. Handle the case that one Op fires multiple kernels. 2. Remove the requirements for CostGraph for easier user adoption, for now. 3. Some speed improvements in graph view. 4. Consolidate the all kinds of tfprof output into one -output option. PiperOrigin-RevId: 155779520 commit 323986c34e5b3467c671f171a3b278340cb597a3 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 12:10:20 2017 -0700 Reduce CPU time and transient heap demand during graph loading. PiperOrigin-RevId: 155779280 commit 912eaee77be9a042578c7566436fd4838b97fd05 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 12:06:19 2017 -0700 Fixing some indenting issues in Markdown. PiperOrigin-RevId: 155778725 commit 999a07105baa551dc84d34a24d45fcbd4894fd20 Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Thu May 11 12:06:00 2017 -0700 Deleted duplicate linkopts keyword commit 5ccf7d3223ef7efdb853c32a26cc417adae2e403 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 11:59:06 2017 -0700 This is running all the benchmark tests on k20, k40, and k80. We had to allow all cpu-platforms for k80. PiperOrigin-RevId: 155777715 commit 6e595abac2f8201e110c6a9facb99c32e6bdb5a3 Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Thu May 11 12:04:30 2017 -0700 Deleted duplicate linkopts keyword commit a4df99a20d13ca1b50ea0eb6d95b9a4d0c4b7cd3 Merge: eb11d6b be15e9e Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Thu May 11 12:02:07 2017 -0700 Merge commit for internal changes commit f2bbf4b9e7d2559ea9667ec791843a19f0f776fc Author: Yutaka Leon <yleon@google.com> Date: Thu May 11 11:56:51 2017 -0700 Move back lookup_ops from feature_column/ to contrib/lookup/ since they are using V1 kernels and we want core use the V2 kernels. PiperOrigin-RevId: 155777403 commit be15e9eb12a2c61f9b0fef98e94967e64af1f6a1 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 11:38:51 2017 -0700 Add option to CTCLoss to skip items with shorter labels than inputs (returns 0-gradient). The default behavior is preserved, namely to return an InvalidArgument which stops training when such an invalid item is encountered. PiperOrigin-RevId: 155774918 commit 3e4c2f6408beaecd3ab3bf8dd199bc5a8d150361 Author: Benoit Steiner <bsteiner@google.com> Date: Thu May 11 11:12:25 2017 -0700 Added a regression test for the graph rewriter options PiperOrigin-RevId: 155771050 commit 31415474123c27197c8553314366425ff2331d36 Author: Alexey Surkov <surkov@google.com> Date: Thu May 11 11:09:58 2017 -0700 Automatically abort stuck transmissions. HttpRequest now uses a progress callback from libcurl to detect and abort transmissions which haven't made any progress for a significant time. PiperOrigin-RevId: 155770703 commit 1f69227b32d9505439aa132667922091be5fca7d Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 10:58:52 2017 -0700 This change reduces the CPU time spent adding nodes to a graph. For an example large graph (13k nodes, 20k edges), this change reduces the CPU time spent loading the graph by 5%. The existing code uses a long sequence of string comparisons and tests, whenever a node is added. The CHECK(class_ == NC_UNITIALIZED) statement can never actually test anything, because all of the string comparisons (except those against empty strings, which serve no purpose) test against a disjoint set of strings, so no collisions are possible. PiperOrigin-RevId: 155768893 commit 8c28f1a2f97858ea949764588d62a02e1ca7a112 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 10:48:25 2017 -0700 Make cmake run.sh executable. PiperOrigin-RevId: 155767313 commit 5407d8951263f8961a6d124b35bbbeb7762253a1 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 10:43:00 2017 -0700 Add a freeze_graph api that can take in GraphDef and SaverDef protos instead of files. PiperOrigin-RevId: 155766519 commit 9df9f6b4dd4af59685e594979aa7f58c14271684 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 10:10:12 2017 -0700 Go: Update generated wrapper functions for TensorFlow ops. PiperOrigin-RevId: 155762112 commit c39029d8de940248eb4610489b53eda0e07a2752 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 10:03:03 2017 -0700 Update ops-related pbtxt files. PiperOrigin-RevId: 155761031 commit 51caa2b03f2975be51ab3f03999f35046b34f4af Author: Chris Song <sjhshy@gmail.com> Date: Thu May 11 18:27:22 2017 +0000 Fix misspells. commit b456315662c84a0721a7db33ddd55ef9d8cde639 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Thu May 11 10:02:12 2017 -0700 Fix expected error text in feature_column_test.py. PiperOrigin-RevId: 155760913 commit 168210843dc8da758f96ca73fc110e4de0d4ce76 Author: Martin Wicke <wicke@google.com> Date: Thu May 11 09:26:17 2017 -0700 Add note clarifying tie breaking behavior. END_PUBLIC BEGIN_PUBLIC Automated g4 rollback of changelist 152465844 PiperOrigin-RevId: 155756924 commit c397ed659f78e5a0fbb2a6099550f80c3ed9ad49 Author: Benoit Steiner <bsteiner@google.com> Date: Thu May 11 08:59:53 2017 -0700 Improved code formatting. PiperOrigin-RevId: 155754247 commit 0f08bba479b290684e215db3eed82a14a61d9270 Author: Petros Mol <pmol@google.com> Date: Thu May 11 08:46:31 2017 -0700 Enables SDCAOptimizer work with dense vectors (of rank 1). By extension, tf.contrib.learn.SVM, tf.contrib.learn.LinearClassifier (when using the SDCA optimizer), SDCALogisticClassifier and SDCALinearRegressor can also work with dense vectors. PiperOrigin-RevId: 155753008 commit 851982c0e3c543306f3942f92bc4bdbc0d50d1f7 Author: Peter Hawkins <phawkins@google.com> Date: Thu May 11 08:30:16 2017 -0700 [TF:XLA] Add function compilation cache. PiperOrigin-RevId: 155751485 commit 1e390f5f7992f3ae5f9ecfc91ebe3d711efe0b7d Author: Benjamin Kramer <kramerb@google.com> Date: Thu May 11 06:58:04 2017 -0700 Add a test for zero-dimensional copies with different layouts, which are nops. PiperOrigin-RevId: 155744287 commit 99706cf26befc2b86606557d01f14b32bf7c73f7 Author: Justine Tunney <jart@google.com> Date: Wed May 10 22:07:14 2017 -0700 Fix build by adding three.d.ts typing PiperOrigin-RevId: 155712533 commit ee112cff56081fb9d0b74c987a8935acc360b05c Author: Benoit Steiner <bsteiner@google.com> Date: Wed May 10 21:12:21 2017 -0700 Merge changes from github. PiperOrigin-RevId: 155709893 commit 27c89207d2f31fe4b4b42c789b96d62cde4e2133 Author: Justine Tunney <jart@google.com> Date: Wed May 10 20:49:53 2017 -0700 Migrate forked TensorBoard build to webfiles bazel run tensorflow/tensorboard/components/tf_tensorboard_d3v4:index This will bring up a local web server and display a vulcanized TensorBoard that is 95% working, with the exception of tf-graph, which isn't included. PiperOrigin-RevId: 155708877 commit 8f8855adc604333180bf57f11e00dc70e679974d Author: Ian Langmore <langmore@google.com> Date: Wed May 10 20:12:11 2017 -0700 Cleanup of a few things in distributions library. Noticed while putting together change for MVNFullCovariance. * Using assert_none_equal * Removing redundant asserts in mvn_tril. PiperOrigin-RevId: 155707155 commit 947fc02a3bfc9bbddac37be2a252c9b7f1b3598b Author: Bjarke Hammersholt Roune <broune@google.com> Date: Wed May 10 19:40:46 2017 -0700 Fold Pad into ReduceWindow. PiperOrigin-RevId: 155705565 commit 326ee1883f873695540e7a85f07f52b24e5974a4 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 19:33:03 2017 -0700 Changes _ReplicaDeviceChooser to not set a task number on ops with a non-PS job. PiperOrigin-RevId: 155705170 commit 9d528b8dc22567e239c0eb692eedb1d9e6763fd1 Author: Justine Tunney <jart@google.com> Date: Wed May 10 18:29:37 2017 -0700 Refactor webfiles rule name to web_library PiperOrigin-RevId: 155701631 commit 6dd1145d57ae3249e1798ca9fe7e81ded5c95ba8 Author: Peter Hawkins <phawkins@google.com> Date: Wed May 10 18:23:47 2017 -0700 [TF:XLA] Make the function library a constructor-time option to XlaCompiler, rather than a CompileFunction()/CompileGraph()-time option. In a subsequent change we plan to add a compilation cache to XlaCompiler, keyed on function name and signature. To avoid incorrect cache hits, we must not use different function libraries in the same XlaCompiler. This change refactors the API so the function library is passed once at construction time, making this invariant more obvious. PiperOrigin-RevId: 155701318 commit 28d2c7b1bf0c6aa7e0f915118fcd4c0be29cbb0e Author: Peter Hawkins <phawkins@google.com> Date: Wed May 10 18:21:23 2017 -0700 [TF:XLA] Add XLA implementation of ResourceApplyAdam. Add a "graph" property to ResourceVariable for compatibility with Variable. PiperOrigin-RevId: 155701128 commit 0e35d4603a8a55e6cc6a2fd979b7b90df9a6dbc4 Author: Benoit Steiner <bsteiner@google.com> Date: Wed May 10 18:15:39 2017 -0700 Added ability to compute SCC for a graphdef. PiperOrigin-RevId: 155700672 commit 79c7dfaca7cd8f119c044db3b1d156dd2be0fad8 Author: Yuefeng Zhou <yuefengz@google.com> Date: Wed May 10 17:55:06 2017 -0700 Decode raw op should return when input is empty. PiperOrigin-RevId: 155699009 commit 5cacaa4ab72acd278c1faa5b16b16a3527136da8 Author: Benoit Steiner <bsteiner@google.com> Date: Wed May 10 17:53:47 2017 -0700 Estimates node completion times for a TF graph based on the op level analytical estimates. PiperOrigin-RevId: 155698917 commit 91cb0e854ba559c7b0ce657e9fa8c4b9644360b7 Author: Justine Tunney <jart@google.com> Date: Wed May 10 17:33:07 2017 -0700 Add Plottable v3 to TensorBoard build This introduces a new macro, which is meant to work around the fact that ES6 imports don't work in the browser. This macro concatenates files, comments out import lines, and adds aliases if necessary. In other words, we're defining the ES6 module structure in the BUILD file, and ignoring what's in the .ts files. PiperOrigin-RevId: 155697059 commit c4c22e756888bc46846a09004e3b4986768da606 Author: Shanqing Cai <cais@google.com> Date: Wed May 10 16:26:10 2017 -0700 tfdbg CLI: add initial support for command "run -p": profiler mode Accompanying changes: * In list_profile / lp output, let the time unit be consistent across all ops, so that the results are easier to read. * Add the --time_unit option flag to list_profile / lp. * Add start time to the lp output table; allow sorting by start time. PiperOrigin-RevId: 155690128 commit e09b0b6ebf6bfea5157a7e8b65b9070e56c1c5f2 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 16:03:02 2017 -0700 Move crossed_column to core. PiperOrigin-RevId: 155687697 commit 770a27161bf9e73860443f62ab539833d39d61b4 Author: Justine Tunney <jart@google.com> Date: Wed May 10 15:47:55 2017 -0700 Introduce one-off solution for vulcanization This is a simple Java script that inlines HTML, CSS, and JavaScript with minification. It tries its best to preserve @license data. The Skylark rule behaves the same as web_library(). It can be bazel run to get the development web server. It outputs the protobuf, because protobuf is awesome. PiperOrigin-RevId: 155686060 commit 2526394fb49f42f2eeb74cf85baecf0787b9f291 Author: Koan-Sin Tan <koansin.tan@gmail.com> Date: Thu May 11 15:13:54 2017 +0800 There is no --logtostderror There is not --logtostderror option. Remove them. And there should be no backslash in --transforms=... commit ab2a1e37c681e9f5f248e3efbaf15a5935c8901a Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 14:07:00 2017 -0700 Merged commit includes the following changes: 155673782 by ashankar <ashankar@google.com>: Windows: Fix bazel builds on Jenkins. PR #9580 added TF_NEED_MKL as an argument to the configure script. So that needs to be set before invoking it on Windows. -- 155671008 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- PiperOrigin-RevId: 155673782 commit f200cbd5229d9e0378fc679153e291fb9ff202f3 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 13:44:07 2017 -0700 Add support for _RealValuedVarLenColumn to {sequence_}input_from_feature_columns. PiperOrigin-RevId: 155670564 commit d61efef9e8c1b0abe369ab725534855d85d96dac Author: Peter Hawkins <phawkins@google.com> Date: Wed May 10 13:34:00 2017 -0700 Add a mechanism for ControlFlowContexts to notify outer contexts of nodes added in an inner context. [TF:XLA] Add checks for Ref types in encapsulate_subgraphs_pass. Add support for overriding the names of the functions generated by encapsulate_subgraphs_pass. Relax an over-zealous check when dumping a graph in function.cc PiperOrigin-RevId: 155669293 commit 01822a426d44b18c311712005e451099761d3823 Author: Jianwei Xie <xiejw@google.com> Date: Wed May 10 13:30:45 2017 -0700 Allows more properties in core RunConfig to be *mutated*. Also improve the usability that if save_checkpoint_steps is mutated, the save_checkpoint_secs should be set as None, and vice verse. Fixes #9795 PiperOrigin-RevId: 155668946 commit 58bc80f083ab78da82bd0fb000e1019d95f1b7cd Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 13:21:13 2017 -0700 Add ability to pass example keys through estimator and return them in the inference dict for TensorForest estimator. PiperOrigin-RevId: 155667839 commit 7e7efbbbc43e46c055c12eb4e7833e9db9e0b18e Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 13:19:39 2017 -0700 Add `weighted_categorical_column`. PiperOrigin-RevId: 155667634 commit bf3e5296f7e2e98a95f260ab797ce2302092fd35 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 13:18:00 2017 -0700 Add utilities to fuse specified nodes as remote fused graph nodes PiperOrigin-RevId: 155667448 commit 1d70e82493f8aee908aea81bed41158d6b9761bf Author: Asim Shankar <ashankar@google.com> Date: Wed May 10 13:17:33 2017 -0700 Java: Allow feeding and fetching colon-separated names. In Python, the string "<operation>:<index>" is referred to as the "tensor name" (see tf.get_tensor_by_name at https://www.tensorflow.org/api_docs/python/tf/Graph#get_tensor_by_name) and this string also appears in the saved model format as the key in the 'inputs' field of the 'SignatureDef' protocol buffer message. Accepting such names makes it easier to use saved models in Java (yes, I know it still isn't easy enough!) For example, using something like: import org.tensorflow.SavedModelBundle; import org.tensorflow.framework.MetaGraphDef; import org.tensorflow.framework.SignatureDef; // ... try (SavedModelBundle model = SavedModelBundle.load(exportDir, "serve")) { final SignatureDef sig = MetaGraphDef.parseFrom(model.metaGraphDef()) .getSignatureDefOrThrow(DEFAULT_SERVING_SIGNATURE_DEF_KEY); model.session().runner() .feed(sig.getInputsOrThrow("inputs").getName()) .fetch(sig.getOutputsOrThrow("outputs").getName()) .run()... } PiperOrigin-RevId: 155667400 commit 15f32d93a48fa78f7f2f17c0a46f5625637c82aa Author: Vijay Vasudevan <vrv@google.com> Date: Wed May 10 12:31:10 2017 -0700 get_attr returns dtype objects instead of raw ints Originally by @Mycosynth, but edited to make it work using integer conversion. Fixes #447. If this breaks you, do an explicit cast of the return of get_attr to int, if you are parsing a dtype object. PiperOrigin-RevId: 155661630 commit 32482dea01e2fc672eee9df5305bccc0b432eb7b Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 11:43:40 2017 -0700 Change soc op type id by input data types and add new hvx ops PiperOrigin-RevId: 155655617 commit b9cdcbe3a019f09eee7e7c9c9039e4533931e1e7 Author: Skye Wanderman-Milne <skyewm@google.com> Date: Wed May 10 11:39:52 2017 -0700 Create a TF_Graph alongside the Python graph. This is a first step towards porting the Python API to use the C API. As the Python Graph and Operations are constructed, an analogous TF_Graph and TF_Operations are created via SWIG. Currently nothing is done with the TF_Graph; a next step will be switching to the new TF_Session API which runs a TF_Graph directly (instead of a GraphDef). This new functionality is disabled by default and can be manually enabled by setting the _USE_C_API global in ops.py. For this patch I only enabled it for a single test file. I tried enabling it for all TF Python tests and manually disabling it for unsupported tests, but there were too many failing tests (although most tests passed). See ops.py for a TODO list of unsupported functionality. I benchmarked building an Inception model, and building the TF_Graph incurs a 20% overhead to the total graph construction time. Note that this patch does not remove any existing Python functionality; another next step will be recovering this time by removing redundant Python code. There is no measurable overhead with the new functionality disabled. PiperOrigin-RevId: 155655064 commit 9472ba05620332fb274e643e08228b5c530e3e52 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 11:25:02 2017 -0700 Automated g4 rollback of changelist 152936284 PiperOrigin-RevId: 155652925 commit 64698dd7bb17dc0796f9635963918059e8a1ebc2 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 10:55:16 2017 -0700 Add `embedding_column`. Add some pydoc to other columns. Add weight_collections tests for all columns. Rename embedding 'weights' -> 'embedding_weights', linear 'weight' -> 'weights', and linear 'bias_weight' -> 'bias_weights'. PiperOrigin-RevId: 155648428 commit 8bda9d06f8a62e995ef14a30d83481874c7720f4 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 10:54:34 2017 -0700 Merged commit includes the following changes: 155648333 by dumitru <dumitru@google.com>: Adding scope argument to slim's DatasetDataProvider. -- 155645726 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal only change. -- PiperOrigin-RevId: 155648333 commit f7eff652ad407d8831e4a05f400f4793673eba2c Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 10:18:16 2017 -0700 Merged commit includes the following changes: 155643058 by cassandrax <cassandrax@google.com>: Add indicator_column (formerly known as one_hot_column) to core TensorFlow. -- 155642205 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- PiperOrigin-RevId: 155643058 commit 39e37da48b20bf46352b92a63d3be7532d043fc6 Author: Geoffrey Irving <geoffreyi@google.com> Date: Wed May 10 09:46:59 2017 -0700 Remove public op_def() access in op_kernel.h This is step 1 towards removing protos from the op definition API. RELNOTES: Remove OpKernelConstruction::op_def() as part of minimizing proto dependencies. PiperOrigin-RevId: 155638924 commit 0ea08a755c27affd3b41abdcd081716fcc9c03ad Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 09:44:52 2017 -0700 Fixes Markdown rendering of UTF-8 strings in text_plugin for Python 2. Markdown does an internal conversion to unicode of any input string passed to it, which fails in case the string contains any non-ASCII characters. This fixes the problem by converting the string to unicode using UTF-8 encoding beforehand. PiperOrigin-RevId: 155638665 commit 8b704af8f767c7483fe16ba3e1a6dc104f8a327b Author: Peter Hawkins <phawkins@google.com> Date: Wed May 10 09:21:42 2017 -0700 [TF:XLA] Split options that pertain to an individual call to XlaCompiler::CompileGraph() or XlaCompiler::CompileFunction() out of XlaCompiler::Options into a separate XlaCompiler::CompileOptions. Remove the 'prune_unreachable_nodes' options since it was unused. PiperOrigin-RevId: 155636247 commit e4f51fc8a2e5bf338159112e4b3dac31bcdf4ae7 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Wed May 10 07:06:12 2017 -0700 [TF:XLA] Optimize the LiteralUtil::Convert() API and some refactoring Optimized the LiteralUtil::Convert() API to use flat scan over the source and destination literal values. Performance boost varies with size. A F32->F64 Convert() of a {32, 64, 64, 64} went from ~35s to ~500ms. Added unittest for LiteralUtil::Convert(). Moved some common/repeated code into a new StrideConfig class. PiperOrigin-RevId: 155624212 commit fbfff7f859ae3d423d0e4ec07d9583261f73fa10 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 21:37:44 2017 -0700 Migrate the tf_graph_dashboard component to use webfiles. Also removed the unused tf_graph_info component. PiperOrigin-RevId: 155591143 commit bca4bb289ba4ae13760f2ab2c6bb19a36743577c Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 20:54:30 2017 -0700 [XLA] Use a templated Visitor for typed evaluation, instead of switching on opcode twice. PiperOrigin-RevId: 155588986 commit 236ef2c9c7b77f851609cb4ee2d82c1428b66c8c Author: Justine Tunney <jart@google.com> Date: Tue May 9 19:22:09 2017 -0700 Make pprof definition deterministic PiperOrigin-RevId: 155584393 commit b808475cc65351f16d383a53547f82f9632aeedd Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 16:32:07 2017 -0700 Remove initialize.h. The REGISTER_MODULE_INITIALIZER macro is used in only one place, and the relevant code is more readable (and not even meaningfully longer) without the macro. PiperOrigin-RevId: 155569181 commit 96980d5f7a8e5f4ae0d06c23663e2f7fecc2e7ba Author: Yuan Yu <yuanbyu@google.com> Date: Tue May 9 16:29:23 2017 -0700 Fix a bug when applying stop_gradient() on the gradient of a while loop. It turned out that I really needed to have tf.gradients() to backprop correctly on the gradient loop even if we have a stop_gradient on it. So, with this CL, I suspect that we are epsilon away to support higher-order gradients for while loops. PiperOrigin-RevId: 155568884 commit 2fd1f880254322e50f65b1739fd502b56074e9f2 Author: Benoit Steiner <bsteiner@google.com> Date: Tue May 9 16:06:49 2017 -0700 Added missing const qualifier PiperOrigin-RevId: 155566351 commit d956e8f6acc9cf9641826e71536ac0d54d6a2d59 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 15:50:20 2017 -0700 Add feature_column.parse_example_spec method. PiperOrigin-RevId: 155564265 commit b661c841c1bb39215661f1b21bd54a4d96ab8ed4 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 15:44:49 2017 -0700 Fix android tests broken by sparse_cross_op. PiperOrigin-RevId: 155563563 commit 7c40495e2eb27b8fe6d213c80c0c9ed37a82607c Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 15:19:06 2017 -0700 Update ops-related pbtxt files. PiperOrigin-RevId: 155560204 commit 05f9de40701ef51b2a21bd8ffd3d1e357b92db68 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 14:51:01 2017 -0700 Capture temporary device tensor, so it doesn't get deallocated before the kernel runs. PiperOrigin-RevId: 155555888 commit c595425a22702ebc1ca18473601414415088caab Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 14:39:40 2017 -0700 Go: Update generated wrapper functions for TensorFlow ops. PiperOrigin-RevId: 155553956 commit 049f9e6a354a1ccc6f1e97064593dede50f07867 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 14:35:36 2017 -0700 Fix formatting of some op descriptions. In most cases, this means separating out a 1-line summary from the main description. PiperOrigin-RevId: 155553359 commit a33e339dd676f370307e67e477141bbacf05cc18 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 14:26:49 2017 -0700 Migrate the tf_graph_board component to use webfiles. PiperOrigin-RevId: 155552095 commit 98469c23b2f289997c69c0f0bebdb35788ce6bb1 Author: Benoit Steiner <bsteiner@google.com> Date: Tue May 9 13:58:56 2017 -0700 Cleaned up the includes of the graph runner. PiperOrigin-RevId: 155548015 commit cc2faf81f988ffcf49269fdc68f27eef07a3ae2c Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 13:00:22 2017 -0700 Automated g4 rollback of changelist 155433842 PiperOrigin-RevId: 155541065 commit 1bad658d63e5fb21d321cd680e7451c96e032f7e Author: Mustafa Ispir <ispir@google.com> Date: Tue May 9 12:32:41 2017 -0700 Moved transform_features. removed pylint lines from test. PiperOrigin-RevId: 155538004 commit 557ab679e730d250d665695d1556d5c2b25f7f07 Author: Asim Shankar <ashankar@google.com> Date: Tue May 9 11:48:18 2017 -0700 Docs: Cosmetic tweak to sample test code. PiperOrigin-RevId: 155532810 commit f6e75f3b78b1ed37f0b4f12a664a88b8cf12dc16 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 10:54:21 2017 -0700 Go: Update generated wrapper functions for TensorFlow ops. PiperOrigin-RevId: 155525165 commit 04d0b083d1b3a663447b815c59ef94e35f3b3ed9 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 10:48:59 2017 -0700 Update ops-related pbtxt files. PiperOrigin-RevId: 155524343 commit d9fc16f61e89b12681b24dc4206370e5a74f4c6f Author: Justin Lebar <jlebar@google.com> Date: Tue May 9 10:28:51 2017 -0700 Unbreak HLO graph dumping for graphs that include a convolution or a C++ template (e.g. "max<float>"). These descriptions include angle brackets, which we need to sanitize out. PiperOrigin-RevId: 155521374 commit 1937df84a17f8f520ab2d45d083d65188afd9c62 Author: Benoit Steiner <bsteiner@google.com> Date: Tue May 9 10:24:22 2017 -0700 Propagate the cluster to the grappler optimizers: this will enable the development of smarter optimizers that can take the available hardware into account when optimizing a graph. PiperOrigin-RevId: 155520653 commit 9a83416501c4a03d6b7f72a046d1b4ad0238cd3b Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 10:22:47 2017 -0700 [TF:XLA] Fix (u)int64 and tensorflow::protobuf_(u)int64 discrepancies Because of the fact that tensorflow::protobuf_uint64 is defined as uint64_t while tensorflow::uint64 is defined as unsigned long long, there exist compiler versions for which the template type matching does not trigger correctly. The LiteralUtil::GetMutableRepeatedField<>() API ends up in having an asymmetric interface because of that, where it needs to be called with tensorflow::protobuf_(u)int64 for 64bit types. This CL removes the LiteralUtil::GetMutableRepeatedField<>() API, confines to Resize<>() the multiple template type expansion, and introduces a new public GetMutableArraySlice<>() API which does not suffer from the discrepancies the LiteralUtil::GetMutableRepeatedField<>() had. PiperOrigin-RevId: 155520416 commit fd85d42d5d8203b4ce17c2825fd003411c78fbe6 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 10:20:55 2017 -0700 Copy sparse_feature_cross to sparse_ops. Two ops will be exposed as sparse_ops._sparse_cross and sparse_ops._sparse_cross_hashed. This cl does not touch the kernels and ops in contrib. PiperOrigin-RevId: 155520054 commit 2ca7c2bdc269b73803d6fa7c199667b987ebeb66 Author: Justin Lebar <jlebar@google.com> Date: Tue May 9 09:54:58 2017 -0700 Add str_util::StringReplace. PiperOrigin-RevId: 155516212 commit 73e537e7826297d2b238ca276db87f7289bb6eb2 Author: Geoffrey Irving <geoffreyi@google.com> Date: Tue May 9 09:16:04 2017 -0700 Show entire shapes in random_crop assertion. Fixes #9708. PiperOrigin-RevId: 155511759 commit 1d0b8c007b8bc7f77dd63c74f02d87185071f038 Author: Peter Hawkins <phawkins@google.com> Date: Tue May 9 09:14:16 2017 -0700 Remove unnecessary copies of value parameters. PiperOrigin-RevId: 155511618 commit b9845c6d0d5dc601fb3b58206a7070aa8937af4f Author: Geoffrey Irving <geoffreyi@google.com> Date: Tue May 9 08:56:07 2017 -0700 Clean up signature documentation for tf.range PiperOrigin-RevId: 155509567 commit 96acbc09ef59bf73c55b8b2dd9556e7967a5f8d8 Author: Geoffrey Irving <geoffreyi@google.com> Date: Tue May 9 08:44:16 2017 -0700 Don't eval in integer softplus test, since it breaks asan PiperOrigin-RevId: 155508401 commit a2ade356755006a2c970d77564e0bb1f70567659 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 08:13:52 2017 -0700 Adding pointer to .tf_configure.bazelrc in the default bazelrc file used by TF containers PiperOrigin-RevId: 155505438 commit b8b9124f2e0c863e57ad96ee17f36e57997e4815 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 02:43:06 2017 -0700 Go: Update generated wrapper functions for TensorFlow ops. PiperOrigin-RevId: 155483248 commit ff0f98bc374bd264560935be5afac0c526be3f00 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 02:35:38 2017 -0700 Update ops-related pbtxt files. PiperOrigin-RevId: 155482813 commit d764056bfeea666898bbec5097ef4a56eedb387c Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 02:30:25 2017 -0700 Fix formatting of some op descriptions by separating out a 1-line summary from the main description. PiperOrigin-RevId: 155482473 commit 90311e78dd6d0a7879bfa37f6aaf0db2e78f0ae9 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue May 9 00:25:31 2017 -0700 Migrate the tf_graph component to use webfiles. This involved rewriting the demo to only use the tf_graph component. We will demo the other components (such as tf-graph-board) in other directories. PiperOrigin-RevId: 155474818 commit 1ece66d58d1dc0ea688248d353b34f44b8ec30ad Author: Justine Tunney <jart@google.com> Date: Mon May 8 23:40:22 2017 -0700 Add more TensorBoard web dependencies to build - Polymer web-component-tester and its deps - weblas for vz-projector - threejs for vz-projector - numericjs for vz-project PiperOrigin-RevId: 155472472 commit 557b421fb0408bfe40f6c6ca37074891e9f97259 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 22:59:22 2017 -0700 Go: Update generated wrapper functions for TensorFlow ops. PiperOrigin-RevId: 155470295 commit 0eff07a37a7dfbcbef5669f4076e9496303671f4 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 22:51:47 2017 -0700 Update ops-related pbtxt files. PiperOrigin-RevId: 155470023 commit 77a65cfe9d1ef561eb70f1554e4df997d7fbff4e Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 22:42:04 2017 -0700 Merged commit includes the following changes: 155469502 by A. Unique TensorFlower <gardener@tensorflow.org>: Fix docs for ConcatOffset op. Can't have anything after arg docs. -- 155468120 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- PiperOrigin-RevId: 155469502 commit 0b783210120d02d8441f87c2f31684b6e4e52349 Author: Sukriti Ramesh <sukritiramesh@google.com> Date: Mon May 8 20:22:01 2017 -0700 SavedModel documentation fix. PiperOrigin-RevId: 155462669 commit dad374b4c70c520405b959950effe1e01cfa658d Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 19:32:53 2017 -0700 Augment compute_constant_test to also test CompileOnlyClient, as well as LocalClient. PiperOrigin-RevId: 155460136 commit bb68623504456430f9c97f3d6118b874cb6dfa0f Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 18:56:10 2017 -0700 Automated g4 rollback of changelist 155428927 PiperOrigin-RevId: 155458050 commit 00f981d1563409c462ea3f69137b0184914b407a Author: Sukriti Ramesh <sukritiramesh@google.com> Date: Mon May 8 18:49:44 2017 -0700 Mark SessionBundle deprecated in documentation. PiperOrigin-RevId: 155457684 commit dc703d80793da9ddd94379b68f859f4195f5540f Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 18:48:00 2017 -0700 Migrate the tf_graph_info component to use webfiles. This involved copying tf-graph-icon.html into the tf_graph_info component (from the tf_graph component). The tf_graph component had not been using tf-graph-icon.html. PiperOrigin-RevId: 155457569 commit 94807f281536d07562f4e281ef1663424cd1c3eb Author: Jianwei Xie <xiejw@google.com> Date: Mon May 8 17:55:20 2017 -0700 Fixed the evaluate call in ValidationMonitor to support core Estimator. PiperOrigin-RevId: 155453312 commit b1820e0d080d3c761b17b63eaf79d0338b2a59f2 Author: Yutaka Leon <yleon@google.com> Date: Mon May 8 17:45:11 2017 -0700 Add initial read-only lookup table ops into core that uses the V2 kernels. PiperOrigin-RevId: 155452370 commit 02f4722a5bcc2ce3be9025112a780d1bc08918f9 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 17:24:51 2017 -0700 These checks only need to be done after the fact if an error occurs while saving. PiperOrigin-RevId: 155450121 commit 6c1f55de2f89ba076c9153a0e33f748f5c468d9a Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 16:24:49 2017 -0700 Adds unit test case for output fused add with non-dot/transpose operand. PiperOrigin-RevId: 155443320 commit 8d254838e35649ce8f6653bfe39758df3eafe43d Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 15:58:29 2017 -0700 Expand "Adding a New Op" to have: - comment about multi-threaded CPU kernel. - example of a GPU kernel. PiperOrigin-RevId: 155440303 commit a57373bd168d4f56bbf6bec86235b4bb40162915 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 15:25:21 2017 -0700 Improve pydoc for `categorical_column_with_identity`. PiperOrigin-RevId: 155435926 commit 960ea0d1633d612d2f8c2f61b4f573f25ce548a3 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 15:09:16 2017 -0700 [XLA] Use a templated Visitor for typed evaluation, instead of switching on opcode twice. PiperOrigin-RevId: 155433842 commit 4b476b8ead27e22b5b90adc103dc4e9ebd7ebce3 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 15:02:59 2017 -0700 Move tf-graph-controls into its own component. Previously, 2 different components tf-graph-app and tf-graph-dashboard had been depending on tf-graph-controls.html, which resided in tf-graph. tf-graph-controls should be its own component because 1. The main tf-graph submodule didn't actually depend on it. Only outside components depended on it. 2. Components like tf-graph-app and tf-graph-dashboard should not really have to depend on HTML files within other components. Also, removed the un-needed <html>, <head>, and <body> tags from the tf_graph_loader demo. PiperOrigin-RevId: 155432856 commit 57a3696510677c54b494a9f04e75eca13ae69d70 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 14:59:32 2017 -0700 Ignore trailing space in transform parameter string. PiperOrigin-RevId: 155432297 commit 036921c833bbe14dd3e21373665e4fd6a7d9f2ad Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 14:51:03 2017 -0700 Small documentation fix: Update code example in cuda_solvers.h after changing method name. PiperOrigin-RevId: 155430968 commit 0a77fc720ba6db07f167b437e12f599ba98fc776 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Tue Feb 7 14:46:29 2017 -0800 Avoid checking for shape consistency in split, when the dimension is unknown. Must defer to runtime. PiperOrigin-RevId: 155430309 commit 8b219918214f779b0f4c7785ae93feffa6e492c3 Author: Shanqing Cai <cais@google.com> Date: Mon May 8 14:46:09 2017 -0700 tfdbg: Internally-oriented changes PiperOrigin-RevId: 155430252 commit 3f1f88c55b22aff9bb900edddbe6cabd4f62a862 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 14:36:28 2017 -0700 Enable USE_GEMM_FOR_CONV and -O3 on android. PiperOrigin-RevId: 155428927 commit c7f79bb75b5b83c3011e164ccd617a6ada910ea4 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 14:30:06 2017 -0700 Merged commit includes the following changes: 155427981 by andrewharp <andrewharp@google.com>: StatSummarizer: Put size check error message outside of if block where it belongs. -- 155427811 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change PiperOrigin-RevId: 155427981 commit 09f3fb939c9b395a9bc747cf81d15b2dc2804c3e Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 14:10:10 2017 -0700 Merged commit includes the following changes: 155425029 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- 155424167 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- PiperOrigin-RevId: 155425029 commit 70c303386909fe1a0d34cada6fe5a42565279849 Author: Mustafa Ispir <ispir@google.com> Date: Mon May 8 13:55:26 2017 -0700 Final cross compatibility cl. Added cross compatibility to contrib feature columns: crossed_column, bucketized_column, scattered_embedding_column. Improved error message in core model builder. PiperOrigin-RevId: 155422877 commit be496fc2d3dbe8146e7174e32fd2c48d2ffe6999 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 13:23:21 2017 -0700 Merged commit includes the following changes: 155418607 by ashankar <ashankar@google.com>: Java: Accessor for list of all registered ops. Adds a method to get all the OpDefs registered in the address space of the process. #9685 #7149 -- 155418502 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change. -- 155418144 by cais <cais@google.com>: Internal-only changes -- PiperOrigin-RevId: 155418607 commit 2de7a9e27ae3b7d6312f8b6842f183c825a80bd7 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 13:19:47 2017 -0700 * Wrap cuBlas high-level solvers into CudaSolver. * Use cuBlas calls to implement a GPU version of the matrix_inverse op. * Refactor CudaSolver to make all the kernel launches asynchronous, and add a public interface "RegisterLapackInfoCheckerCallback", which registers a callback to be invoked when the statuses of one or more solver kernels passed to it have been copied back to the host. * Refactor CholeskyOpGpu to use the updated API and make that kernel async too. PiperOrigin-RevId: 155418113 commit 50a22848cef10e924a5b72fd0ca2741d8366161b Author: Dan Ringwalt <ringwalt@google.com> Date: Mon May 8 13:19:24 2017 -0700 Support a list of globs in file_io.get_matching_files. PiperOrigin-RevId: 155418052 commit 3be35ada0ee4984d57aa3a1a887f778227ffc4c9 Author: Ian Langmore <langmore@google.com> Date: Mon May 8 13:17:37 2017 -0700 VectorLaplaceDiag distribution added. VectorLaplaceLinearOperator added but not exposed as part of API. .det_covariance and .log_det_covariance removed from MVNLinOp Some small docfix to MVN PiperOrigin-RevId: 155417807 commit 1ea40e1ef95e28c702a26a31d69879edc0adbcee Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 13:14:45 2017 -0700 Add nomac filter to mac build. PiperOrigin-RevId: 155417322 commit 362e441b6d1dbcdfed02e0251a8f40adfc0a9d66 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 12:58:57 2017 -0700 Add an option to remove empty rows and columns from the input matrix, in the input function used in the test for the WALSMatrixFactorization estimator. PiperOrigin-RevId: 155415303 commit 10096f75722e56eddb95512532842e3082b23cfc Author: Justin Lebar <jlebar@google.com> Date: Mon May 8 12:25:42 2017 -0700 Beautify HLO dot output. - Show the value of effective scalar constants. Previously we only showed the values of true (i.e., R0) scalars. - Change the graph's font to Roboto. Previously it was...whatever the default is -- Times New Roman? - Place the graph's heading at the top of the graph, instead of wherever the dot renderer pleases. - Make all nodes other than "while" rectangles, and Use rectangles with rounded corners for nested computations, to make them easier to distinguish from the other nodes. - Make nodes' opcodes and graph headings bold. - Tweak the graph's colors. * Now we use Material Design colors, from https://material.io/color. These are in general much lighter, which I think lets the reader focus more on the nodes' contents. * tuple and get-tuple-element are now white, to reflect that they are usually nops. This makes graphs with many GTE ops much easier to parse. (I think this is the most impactful change in this patch.) * We use fewer unique colors now, again in an effort to help readers focus on the graph itself. This necessitates making some ops that previously had different colors share colors: - send, recv, infeed, outfeed, and cross-replica-sum are all now the same color, since they represent data transfer. - dot and conv are now both the same (bold) color, since these are often the most expensive ops in models. - map and fusion are now the same color, since they both are wrappers around "lambda functions". PiperOrigin-RevId: 155411313 commit ebd6de64299a5484265455d8e0f469fad91a0927 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 12:18:14 2017 -0700 Fix flaky test batch_sequences_with_states PiperOrigin-RevId: 155410358 commit a816db6b977fac2e3ff6ccd490529aa558f1a9d1 Author: Jianwei Xie <xiejw@google.com> Date: Mon May 8 12:17:18 2017 -0700 Fixes typos in docstring of string table. PiperOrigin-RevId: 155410244 commit eaa18b3d0fc4bd71d9b36a55fda81f7787427e8f Author: Shanqing Cai <cais@google.com> Date: Mon May 8 12:09:39 2017 -0700 Actually enforce nomac tag on Mac Previously the nomac tag was not enforced, perhaps by mistake. This CL fixes that. PiperOrigin-RevId: 155409422 commit 41f86fd4220e3e92b756c2c18033c693655baea3 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 12:04:35 2017 -0700 Fixed typo. PiperOrigin-RevId: 155408790 commit 5dd25a29433cdfb9938f6766fd53ffc5f05972c9 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 11:56:19 2017 -0700 Merged commit includes the following changes: 155407708 by yaozhang <yaozhang@google.com>: Only search for dequeue node in the transitive fanin of the training outputs. For some models, there are multiple dequeue nodes respectively for training, validation, and test. However, only the training dequeue node is of our interest. -- 155407641 by A. Unique TensorFlower <gardener@tensorflow.org>: Internal change -- PiperOrigin-RevId: 155407708 commit 1244c8fdda9efb11a5386dce40fbb885f0df190b Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 11:39:22 2017 -0700 This is to add the tf.contrib.image.bipartite_match op. It can find bipartite matching based on a given distance matrix. A greedy bi-partite matching alogrithm is used to obtain the matching with the (greedy) minimum distance. PiperOrigin-RevId: 155405406 commit e884e63f66d9d5c80df4a801dd94300aefbe7127 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 11:18:56 2017 -0700 Add mac testing script. PiperOrigin-RevId: 155402284 commit 3bdeb9f8ce91c3234320cd559b6ae5ac19394981 Author: Jianwei Xie <xiejw@google.com> Date: Mon May 8 11:14:32 2017 -0700 Adds replace api into RunConfig. PiperOrigin-RevId: 155401586 commit eb11d6bf1b3e2aee52ff52dc03d07fbf83f7ae3b Merge: da3c408 78fe194 Author: Yifei Feng <fengyifei2026@gmail.com> Date: Wed May 10 14:02:12 2017 -0700 Merge pull request #9776 from benoitsteiner/branch_155393864 Branch 155393864 commit da3c4080ae7c0a5820977f2af46d6996c0d1838f Merge: 7fa0cf3 16271ea Author: Yifei Feng <fengyifei2026@gmail.com> Date: Wed May 10 13:18:46 2017 -0700 Merge pull request #9452 from zasdfgbnm/flatten add support for flat both inner and outer dims commit 11a174e6f6a865b4f046910e5f7f113cece2eeaa Merge: 924b057 7fa0cf3 Author: Jeroen Bédorf <jeroen@minds.ai> Date: Mon May 8 22:04:56 2017 -0700 Merge branch 'master' of https://github.com/tensorflow/tensorflow into MPIV4 commit 924b057f17fd96141ca5dc3462b2893d28ea64d8 Author: Jeroen Bédorf <jeroen@minds.ai> Date: Mon May 8 21:58:57 2017 -0700 Type fixes commit 92581f0ae06326e88a5e878334055d26979e95ba Author: Jeroen Bédorf <jeroen@minds.ai> Date: Mon May 8 21:48:20 2017 -0700 README added commit 78fe1944d9af6f5826be94679269558a8c6437a5 Merge: 7fa0cf3 697f34c Author: Benoit Steiner <benoit.steiner.goog@gmail.com> Date: Mon May 8 19:31:28 2017 -0700 Merge commit for internal changes commit 697f34ca82c4488f3bb2d3b1e930b0088f63f8c5 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 09:20:04 2017 -0800 Extend tfprof to associate op stats with Python codes. It's backward compatible. Stats of a source code line are aggregated from all ops created by that line. A example. _TFProfRoot (0us/22.44ms) model_analyzer_test.py:149:run_filename_as_m...:none (0us/22.44ms) model_analyzer_test.py:33:_run_code_in_main:none (0us/22.44ms) model_analyzer_test.py:208:<module>:test.main() (0us/22.44ms) model_analyzer_test.py:132:testComplexCodeView:x = lib.BuildFull... (0us/22.44ms) model_analyzer_testlib.py:63:BuildFullModel:return sgd_op.min... (0us/21.83ms) model_analyzer_testlib.py:54:BuildFullModel:seq.append(array_... (0us/254us) model_analyzer_testlib.py:42:BuildSmallModel:x = nn_ops.conv2d... (0us/134us) ... model_analyzer_testlib.py:61:BuildFullModel:loss = nn_ops.l2_... (0us/28us) model_analyzer_test.py:134:testComplexCodeView:sess.run(variable... (0us/0us) Change: 155393864 commit ec8ffb9eafd98b65ed0b594e546c49af89c298c5 Author: Mustafa Ispir <ispir@google.com> Date: Mon May 8 08:46:55 2017 -0800 Improved documentation of bucketized_column. Change: 155389503 commit dcb6e92b1d0833ad9f30536896f5874260f80995 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 08:46:22 2017 -0800 [XLA] Service::CreateModuleConfig should take the backend as a parameter, rather than assuming it's creating a module for the execute backend. This fixes the TODO from service.cc that I added in my previous commit. Change: 155389451 commit e6bf33e8eb62a9490b9621316fe757c3f500c6f3 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Mon May 8 06:29:01 2017 -0800 Added missing dependencies in llvm.BUILD. Change: 155376589 commit d592502460f5647422cf490d44243a8f5a41a882 Author: Eric Liu <ioeric@google.com> Date: Mon May 8 00:37:29 2017 -0800 Add plugin_asset_util.PluginDirectory that returns plugin's logdir. Change: 155354289 commit 3c02d1100788789b04e04feb93761f0ad898ea77 Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Sun May 7 22:47:19 2017 -0800 In the CUDA path of depthwise_conv2d, add __launch_bounds__ and use CUDA runtime helpers to determine LaunchConfig which maximizes occupancy. Change: 155347912 commit f7935b8f8e4d51802ca7eff1c9f740446d6ca24a Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Sun May 7 16:07:06 2017 -0800 Migrate the tf_graph_loader component to use webfiles. Change: 155334410 commit 60adca7bce76c76cab2f872426cecf4b63a64526 Author: David Majnemer <majnemer@google.com> Date: Sun May 7 14:18:47 2017 -0800 [XLA] Make LiteralUtil::EachCellAsString handle all ranks LiteralUtil::EachCellAsString was hard coded to handle only up to rank 4. This simply reworks it to handle all ranks. Change: 155331612 commit 4dc3fdca610b567628e7b1dfd19bd6eaa7238733 Author: Petros Mol <pmol@google.com> Date: Sun May 7 11:22:29 2017 -0800 Have sdca_model_fn used in SDCA-based optimizers use the training Ops from SDCAOptimizer instead of replicating this logic in place. Change: 155326885 commit 7fa0cf39f854d5fdaaa19ad6425dfed02f5fea64 Author: Zader Zheng <yumaoshu@gmail.com> Date: Mon May 8 08:31:50 2017 +0800 Unveil type check part in _VerifyGeneratedGradients that was not actu… (#9529) * unveil type check part in _VerifyGeneratedGradients that was not actually used * remove the block that's not getting used commit bdca561c273d3749a8359763c157daa850af138c Author: Jianfei Wang <me@thinxer.com> Date: Mon May 8 02:02:58 2017 +0800 Fix tf.meshgrid documentation (#9740) The original document has the wrong output regarding the default indexing xy. x = [1, 2, 3] y = [4, 5, 6] X, Y = tf.meshgrid(x, y) print(X.eval()) print(Y.eval()) [[1 2 3] [1 2 3] [1 2 3]] [[4 4 4] [5 5 5] [6 6 6]] commit d1565250256215a5e8bea184f3848b9d468672cb Author: Mortada Mehyar <mortada@users.noreply.github.com> Date: Sat May 6 23:22:17 2017 -0700 Fix missing dash for cxxopt build option (#9736) commit 6ee900438cb0e25b082f7ab54f677b07ce812ebd Author: Bairen Yi <byronyi@users.noreply.github.com> Date: Sun May 7 14:21:37 2017 +0800 fix doc (#9733) commit 9d7e9b0e6b6a08cf1613f073d1459d7ba42034fe Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Sat May 6 21:11:19 2017 -0800 [TF:XLA] Instead of dumping ptxas output, write to a temp file & and use to build gpu executable. Switching to ptxas to avoid relying on the JIT in the nvidia driver for compilation. Change: 155305963 commit 1968b8b3f9316a8327176e25211c8cc2b01459bc Author: A. Unique TensorFlower <gardener@tensorflow.org> Date: Sat May 6 20:46:07 2017 -0800 Only make 1 collapsible for each run. Previously, we could possibly make several collapsibles for each run because of how tag names were broken up. Change: 155305237 commit 196f63b0fd2b4564b4be147ad305224ec465f0df Merge: 3358855 739718f Author: Shanqing Cai <cais@google.com> Date: Sat May 6 23:52:47 2017 -0400 Merge pull request #9732 from jhjin/master better handling of MKL check in configure commit 24a61445f419c9698337cd40346b943b682e739f Author: Yuefeng Zhou <yuefengz@google.com> Date: Sat May 6 18:45:33 2017 -0800 track persistent memory for temporary memory op. Change: 155302379 commit 93572de9a17ca687318c7afac4496f515cd2264d Author: Anna R <annarev@google.com> Date: Sat May 6 18:07:31 2017 -0800 Internal change. Change: 155301612 commit 7cac7f24d1c9c80d8ff8b17d4a52605486f1550f Author: Yao Zhang <yaozhang@google.com> Date: Sat May 6 14:11:14 2017 -0800 Specify a different outer class name for the generation of java output. Change: 155296402 commit 16271ea2fb61319554cecbd1be773a44b89d2de2 Author: Xiang Gao <qasdfgtyuiop@gmail.com> Date: Sat May 6 17:56:54 2017 -0400 pass tests commit 739718f95ce114b1ffceeb23804f119c9b3d3474 Author: Jonghoon Jin <jhjin0@gmail.com> Date: Sat May 6 12:51:14 2017 -0700 better handling of MKL check in configure not prompted to ask MKL download if the var is set already remove unexpected exit when input is given by user commit 10f625166986331521232a692a5e3a8fc933c1c4 Merge: 330a687 3358855 Author: Xiang Gao <qasdfgtyuiop@gmail.com> Date: Sat May 6 17:22:30 2017 -0400 Merge remote-tracking branch 'upstream/master' into flatten commit 330a6873ece3516f883e5e644af5eed66c9c7ac7 Merge: 629fc4e a362e0d Author: Xiang Gao <qasdfgtyuiop@gmail.com> Date: Sat May 6 17:21:33 2017 -0400 …
This is a feature request, following a change in behavior from TF1.0 to TF1.1.
my_estimator = tf.estimator.Estimator(
model_fn = my_model_fn,
model_dir = my_model_dir,
config=tf.contrib.learn.RunConfig(
save_checkpoints_steps=20,
save_checkpoints_secs=None,
save_summary_steps=40,
)
)
Describe the problem
on tf version 1.0 we could easily configure the checkpoint dump (and more) of tf.contrib.learn.Estimator, as stated above: by sending as an input to the estimator tf.contrib.learn.RunConfig with the desired configuration.
on tf version 1.1 this became more complex: the above (and more) configurations in RunConfig are static, and changing them requires either changing the original tf code, or maybe creating a class inheriting from tf.estimator.RunConfig to serve as the config for the estimator.
Source code / logs
in tf.estimator.RunConfig (v1.1), e.g.:
@Property
def save_checkpoints_secs(self):
return 600
in tf.contrib.learn.RunConfig (v1.0, yet supported on tf1.1):
@Property
def save_checkpoints_secs(self):
return self._save_checkpoints_secs
The text was updated successfully, but these errors were encountered: