platform/upstream/tensorflow.git
6 years agoGo: Update generated wrapper functions for TensorFlow ops.
A. Unique TensorFlower [Thu, 12 Apr 2018 02:45:42 +0000 (19:45 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId: 192548367

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Thu, 12 Apr 2018 02:18:20 +0000 (19:18 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 192546579

6 years ago[XLA] Redesign: fix GetComputationGraphStats.
A. Unique TensorFlower [Thu, 12 Apr 2018 02:14:47 +0000 (19:14 -0700)]
[XLA] Redesign: fix GetComputationGraphStats.

CreateFromProto requires that the config has proper entry_computation_layout, so give the config the program shape.

PiperOrigin-RevId: 192546316

6 years agoSwitch to WaitForNotification to fix the flaky test.
Jonathan Hseu [Thu, 12 Apr 2018 01:57:49 +0000 (18:57 -0700)]
Switch to WaitForNotification to fix the flaky test.

See:
https://source.cloud.google.com/results/invocations/31632a30-3728-4635-a456-f89b9e8b9dfe/log
PiperOrigin-RevId: 192544848

6 years ago[XLA] Redesign: test sharding.
A. Unique TensorFlower [Thu, 12 Apr 2018 01:38:38 +0000 (18:38 -0700)]
[XLA] Redesign: test sharding.

Also set the sharding to the instruction when created from proto.

PiperOrigin-RevId: 192543024

6 years agoIncrease size of //tensorflow/python/kernel_tests:linalg_ops_test to "medium".
Derek Murray [Thu, 12 Apr 2018 01:37:47 +0000 (18:37 -0700)]
Increase size of //tensorflow/python/kernel_tests:linalg_ops_test to "medium".

PiperOrigin-RevId: 192542956

6 years agoIn model_to_estimator, only run get_weights when there are initialized Keras variable...
Yifei Feng [Thu, 12 Apr 2018 01:20:19 +0000 (18:20 -0700)]
In model_to_estimator, only run get_weights when there are initialized Keras variables(which assumes there exists a session). Otherwise create a session so that we can run get_config().
Actually fix #18193.

PiperOrigin-RevId: 192541442

6 years agoAdd `tf.contrib.stateless.stateless_multinomial()`.
Derek Murray [Thu, 12 Apr 2018 01:09:42 +0000 (18:09 -0700)]
Add `tf.contrib.stateless.stateless_multinomial()`.

This is a starting point for Dataset-compatible weighted sampling across a list of datasets.

PiperOrigin-RevId: 192540412

6 years agoInternal testing changes
Austin Anderson [Thu, 12 Apr 2018 00:46:08 +0000 (17:46 -0700)]
Internal testing changes

PiperOrigin-RevId: 192537874

6 years agoAutomated g4 rollback of changelist 192516190
A. Unique TensorFlower [Thu, 12 Apr 2018 00:29:32 +0000 (17:29 -0700)]
Automated g4 rollback of changelist 192516190

PiperOrigin-RevId: 192536085

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Thu, 12 Apr 2018 00:19:20 +0000 (17:19 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 192534931

6 years agoboosted_trees: make sure ensemble deserialization happens for the non-TRAIN modes...
Younghee Kwon [Wed, 11 Apr 2018 23:59:45 +0000 (16:59 -0700)]
boosted_trees: make sure ensemble deserialization happens for the non-TRAIN modes too.

PiperOrigin-RevId: 192532297

6 years agoMove callback into bound function to avoid copying.
Derek Murray [Wed, 11 Apr 2018 23:43:33 +0000 (16:43 -0700)]
Move callback into bound function to avoid copying.

PiperOrigin-RevId: 192530231

6 years agoSpeed up computing mean confidence intervals by avoiding tf.while_loop.
A. Unique TensorFlower [Wed, 11 Apr 2018 23:05:42 +0000 (16:05 -0700)]
Speed up computing mean confidence intervals by avoiding tf.while_loop.

Implement a vectorized way to compute the same thing instead.

PiperOrigin-RevId: 192524667

6 years agosupport profiling multiple tpu through one grpc and one session.
A. Unique TensorFlower [Wed, 11 Apr 2018 22:59:47 +0000 (15:59 -0700)]
support profiling multiple tpu through one grpc and one session.
data are saved with host prefix.

PiperOrigin-RevId: 192523668

6 years agoAdding support of core feature columns and losses to gradient boosted trees estimators.
A. Unique TensorFlower [Wed, 11 Apr 2018 22:44:55 +0000 (15:44 -0700)]
Adding support of core feature columns and losses to gradient boosted trees estimators.

PiperOrigin-RevId: 192521398

6 years agoRevealing the range of node ids in the latest layer via resource' state
A. Unique TensorFlower [Wed, 11 Apr 2018 22:37:49 +0000 (15:37 -0700)]
Revealing the range of node ids in the latest layer via resource' state

PiperOrigin-RevId: 192520351

6 years agoIncrease size of //tensorflow/python/kernel_tests:sets_test to "medium".
Derek Murray [Wed, 11 Apr 2018 22:33:06 +0000 (15:33 -0700)]
Increase size of //tensorflow/python/kernel_tests:sets_test to "medium".

PiperOrigin-RevId: 192519639

6 years agoFixes issue where name scope collisions could lead to an invalid variable in the...
Alexandre Passos [Wed, 11 Apr 2018 22:23:17 +0000 (15:23 -0700)]
Fixes issue where name scope collisions could lead to an invalid variable in the metagraph.

PiperOrigin-RevId: 192518307

6 years agoGCS Filesystem should not cache checkpoint file as we need to read the updated checkp...
Sourabh Bajaj [Wed, 11 Apr 2018 22:20:11 +0000 (15:20 -0700)]
GCS Filesystem should not cache checkpoint file as we need to read the updated checkpoints from the contents.

PiperOrigin-RevId: 192517819

6 years agoExtend support to remove transpose/reverse on dimensions of size 1.
A. Unique TensorFlower [Wed, 11 Apr 2018 22:09:07 +0000 (15:09 -0700)]
Extend support to remove transpose/reverse on dimensions of size 1.

PiperOrigin-RevId: 192516190

6 years agoAdding support for batch_to_space_nd op with crops.
A. Unique TensorFlower [Wed, 11 Apr 2018 21:39:11 +0000 (14:39 -0700)]
Adding support for batch_to_space_nd op with crops.

PiperOrigin-RevId: 192511036

6 years agoAdds a nodedef_fn parameter to copy_op_handler, allowing customization by mutating
A. Unique TensorFlower [Wed, 11 Apr 2018 21:04:09 +0000 (14:04 -0700)]
Adds a nodedef_fn parameter to copy_op_handler, allowing customization by mutating
NodeDef before creating the copied operation.

PiperOrigin-RevId: 192505209

6 years agoBREAKING_CHANGE: Remove event_ndims in Bijector, and require `log_det_jacobian` metho...
A. Unique TensorFlower [Wed, 11 Apr 2018 21:02:49 +0000 (14:02 -0700)]
BREAKING_CHANGE: Remove event_ndims in Bijector, and require `log_det_jacobian` methods to take event_ndims.

The class level event_ndims parameter is being deprecated in favor of passing it in
to the `log_det_jacobian` methods.

Specific changes:

  - `log_det_jacobian` signatures are now `log_det_jacobian(input, event_ndims)`

  - Constructors no long have event_ndims passed in (e.g. Affine() vs. Affine(event_ndims=0)).

  - All bijectors must specify a subset of [forward_min_event_ndims, inverse_min_event_ndims]. This is the minimal dimensionality the bijector operates on, with it being "broadcasted" to any passed in event_ndims (e.g. Exp has forward_min_event_ndims = 0. That means it operates on scalars. However, we can use the bijector on any event_ndims > 0 (i.e. we've broadcasted the transformation to work on any amount of event_ndims > 0), and jacobian reduction will work in those cases.

As a result of this change, all bijectors should "broadcast" (e.g. Sigmoid now works on any number of event_ndims).

Other changes (internal and documentation):
  - Added clarifications on Jacobian Determinant vs. Jacobian Matrix.
  - Added clarifications on min_event_ndims, and what the jacobian reduction is over.
  - Changed caching of ildj to be keyed on event_ndims.
  - Several bug fixes to bugs unearthed while writing this code (e.g. transformed distribution shape computation being incorrect)

PiperOrigin-RevId: 192504919

6 years agoPorting tests for the `decode_proto` and `encode_proto` to OS.
Jiri Simsa [Wed, 11 Apr 2018 20:59:58 +0000 (13:59 -0700)]
Porting tests for the `decode_proto` and `encode_proto` to OS.

PiperOrigin-RevId: 192504411

6 years agoSupport for removing unfused quantized activation functions and min/max.
A. Unique TensorFlower [Wed, 11 Apr 2018 20:53:01 +0000 (13:53 -0700)]
Support for removing unfused quantized activation functions and min/max.

PiperOrigin-RevId: 192503204

6 years agoReplace examples/image_retraining by a pointer to TensorFlow Hub.
A. Unique TensorFlower [Wed, 11 Apr 2018 20:48:43 +0000 (13:48 -0700)]
Replace examples/image_retraining by a pointer to TensorFlow Hub.

https://github.com/tensorflow/hub/tree/master/examples/image_retraining
has the same tool, upgraded to use TensorFlow Hub instead of raw graph defs.

PiperOrigin-RevId: 192502469

6 years agoFactor out the syntactic function scope tracking into the transformer. Choosing not...
A. Unique TensorFlower [Wed, 11 Apr 2018 20:47:46 +0000 (13:47 -0700)]
Factor out the syntactic function scope tracking into the transformer. Choosing not to do this at static analysis because it exposes the scope to any node, making it easier to use by any specialization of a transformer.

PiperOrigin-RevId: 192502309

6 years agoWhen not necessary, avoid the creation of a `placeholder_with_default` in BN (not...
Francois Chollet [Wed, 11 Apr 2018 20:46:03 +0000 (13:46 -0700)]
When not necessary, avoid the creation of a `placeholder_with_default` in BN (not yet supported by TPU compilation).

PiperOrigin-RevId: 192502020

6 years agoDepthwiseConv Optimizations
A. Unique TensorFlower [Wed, 11 Apr 2018 20:29:12 +0000 (13:29 -0700)]
DepthwiseConv Optimizations

PiperOrigin-RevId: 192499401

6 years agoInternal change.
Anna R [Wed, 11 Apr 2018 20:22:53 +0000 (13:22 -0700)]
Internal change.

PiperOrigin-RevId: 192498471

6 years agoAdding hp5y back.
Rohan Jain [Wed, 11 Apr 2018 19:33:04 +0000 (12:33 -0700)]
Adding hp5y back.

PiperOrigin-RevId: 192491335

6 years agoInternal TF Lite test changes
Austin Anderson [Wed, 11 Apr 2018 19:32:08 +0000 (12:32 -0700)]
Internal TF Lite test changes

PiperOrigin-RevId: 192491201

6 years agoIncrease size of tensorflow/contrib/data/python/kernel_tests:batch_dataset_op_test...
Derek Murray [Wed, 11 Apr 2018 18:52:52 +0000 (11:52 -0700)]
Increase size of tensorflow/contrib/data/python/kernel_tests:batch_dataset_op_test to "medium".

PiperOrigin-RevId: 192484895

6 years agoUpdate docs for softmax_cross_entropy_with_logits.
Adria Puigdomenech [Wed, 11 Apr 2018 18:36:56 +0000 (11:36 -0700)]
Update docs for softmax_cross_entropy_with_logits.

PiperOrigin-RevId: 192482242

6 years agoScript to create custom_ops inside a TensorFlow graphdef.
A. Unique TensorFlower [Wed, 11 Apr 2018 18:33:38 +0000 (11:33 -0700)]
Script to create custom_ops inside a TensorFlow graphdef.

PiperOrigin-RevId: 192481690

6 years agoDisable prelu tests for real now.
A. Unique TensorFlower [Wed, 11 Apr 2018 18:26:25 +0000 (11:26 -0700)]
Disable prelu tests for real now.

PiperOrigin-RevId: 192480452

6 years agoFix uninitialized value.
A. Unique TensorFlower [Wed, 11 Apr 2018 18:21:48 +0000 (11:21 -0700)]
Fix uninitialized value.

PiperOrigin-RevId: 192479630

6 years agoUse tf.train.get_or_create_global_step() instead of deprecated variables.get_or_creat...
A. Unique TensorFlower [Wed, 11 Apr 2018 18:02:48 +0000 (11:02 -0700)]
Use tf.train.get_or_create_global_step() instead of deprecated variables.get_or_create_global_step().

PiperOrigin-RevId: 192476077

6 years agoAdd gradient in cond test to match CallGradInLoop.
Jacques Pienaar [Wed, 11 Apr 2018 16:53:21 +0000 (09:53 -0700)]
Add gradient in cond test to match CallGradInLoop.

PiperOrigin-RevId: 192463997

6 years agoAdd a clear error message for when a doc does not have a title.
Mark Daoust [Wed, 11 Apr 2018 16:50:40 +0000 (09:50 -0700)]
Add a clear error message for when a doc does not have a title.

PiperOrigin-RevId: 192463583

6 years agoImport FunctionDef as GrapplerFunctionItem
A. Unique TensorFlower [Wed, 11 Apr 2018 16:43:32 +0000 (09:43 -0700)]
Import FunctionDef as GrapplerFunctionItem

Explicitly track function input arg expansion into Placeholders,
and keep metadata to map between FunctionDef and GraphDef connectivity
formats.

PiperOrigin-RevId: 192462592

6 years agoFixing dependencies.
Nupur Garg [Wed, 11 Apr 2018 16:34:51 +0000 (09:34 -0700)]
Fixing dependencies.

PiperOrigin-RevId: 192461382

6 years agoFix Windows GPU TensorFlow Bazel builds.
Michael Case [Wed, 11 Apr 2018 16:34:44 +0000 (09:34 -0700)]
Fix Windows GPU TensorFlow Bazel builds.

The configure.py script will error out on Windows GPU builds due
to NCCL attempted to be configured (and is currently Linux only).

PiperOrigin-RevId: 192461362

6 years agoTemporarily remove prelu from generated_examples_zip_test
A. Unique TensorFlower [Wed, 11 Apr 2018 15:30:18 +0000 (08:30 -0700)]
Temporarily remove prelu from generated_examples_zip_test

PiperOrigin-RevId: 192453411

6 years agoRemove unused former source of tensorflow.org/tutorials/image_retraining.
A. Unique TensorFlower [Wed, 11 Apr 2018 15:11:50 +0000 (08:11 -0700)]
Remove unused former source of tensorflow.org/tutorials/image_retraining.

The source of https://tensorflow.org/tutorials/image_retraining has moved from
https://github.com/tensorflow/tensorflow/tree/master/tensorflow/docs_src/tutorials
to https://github.com/tensorflow/hub/tree/master/docs/tutorials
because of its use of TensorFlow Hub.

This change replaces the now-defunct version with a pointer to the new
location, in order to avoid dead code.

PiperOrigin-RevId: 192451570

6 years agoFixing propagation of minmax info on constant gather ops.
A. Unique TensorFlower [Wed, 11 Apr 2018 14:47:26 +0000 (07:47 -0700)]
Fixing propagation of minmax info on constant gather ops.

PiperOrigin-RevId: 192448922

6 years ago[TF:XLA] Mark oom_test as optonly, it's really slow when compiled without optimization.
Benjamin Kramer [Wed, 11 Apr 2018 09:16:25 +0000 (02:16 -0700)]
[TF:XLA] Mark oom_test as optonly, it's really slow when compiled without optimization.

PiperOrigin-RevId: 192420481

6 years agoCreate FileWriter <-> tf.contrib.summary compatibility layer
Nick Felt [Wed, 11 Apr 2018 06:44:12 +0000 (23:44 -0700)]
Create FileWriter <-> tf.contrib.summary compatibility layer

This provides an implementation of FileWriter, activated by passing in a `session` parameter to the constructor, that is backed by session.run'ing graph ops that manipulate a tf.contrib.summary.create_file_writer() instance. Because tf.contrib.summary.SummaryWriters are backed by shared resources in the graph, this makes it possible to have a FileWriter and a tf.contrib.summary.SummaryWriter that both write to the same events file.

This change includes some related smaller changes:
- Factors out training_utils.py into a separate target to avoid a cyclic dep
- Moves contrib/summary/summary_ops.py to python/ops/summary_ops_v2.py
- Adds SummaryWriter.init(), .flush(), and .close() op-returning methods
- Changes create_file_writer() `name` arg to default to logdir prefixed by `logdir:` so shared resources are scoped by logdir by default
- Fixes a bug with tf.contrib.summary.flush() `writer` arg
- Makes create_file_writer()'s max_queue arg behave as documented
- Adds more testing for existing tf.contrib.summary API

PiperOrigin-RevId: 192408079

6 years ago[XLA] Fix the size of data buffer for sparse literals.
A. Unique TensorFlower [Wed, 11 Apr 2018 05:44:36 +0000 (22:44 -0700)]
[XLA] Fix the size of data buffer for sparse literals.

PiperOrigin-RevId: 192404543

6 years ago[XLA] Redesign: implement ComputeHost.
A. Unique TensorFlower [Wed, 11 Apr 2018 05:29:13 +0000 (22:29 -0700)]
[XLA] Redesign: implement ComputeHost.

Also support convert from/to proto for ComputeHost.

PiperOrigin-RevId: 192403660

6 years ago[XLA] Redesgin: implement and test Gather, Conditional.
A. Unique TensorFlower [Wed, 11 Apr 2018 04:42:14 +0000 (21:42 -0700)]
[XLA] Redesgin: implement and test Gather, Conditional.

Also support convert from/to proto for Gather.

PiperOrigin-RevId: 192400659

6 years ago[XLA] Redesign: implment and test CrossReplicaSum.
A. Unique TensorFlower [Wed, 11 Apr 2018 03:48:57 +0000 (20:48 -0700)]
[XLA] Redesign: implment and test CrossReplicaSum.

PiperOrigin-RevId: 192397189

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Wed, 11 Apr 2018 02:44:00 +0000 (19:44 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 192392702

6 years ago[StreamExecutor] Merge StreamExecutor's and XLA's StatusOr classes.
Justin Lebar [Wed, 11 Apr 2018 02:34:54 +0000 (19:34 -0700)]
[StreamExecutor] Merge StreamExecutor's and XLA's StatusOr classes.

StatusOr is a...complicated class to write.  It's really not good to
have two copies of it.  They've diverged (the XLA one is more
sophisticated), and this may be causing upstream build problems with
gcc6.

PiperOrigin-RevId: 192392111

6 years ago[XLA] Redesign: implement and test BatchNormXXX.
A. Unique TensorFlower [Wed, 11 Apr 2018 02:31:05 +0000 (19:31 -0700)]
[XLA] Redesign: implement and test BatchNormXXX.

PiperOrigin-RevId: 192391748

6 years agoexperimental C API: Fix compilation failure in Windows.
Asim Shankar [Wed, 11 Apr 2018 02:20:58 +0000 (19:20 -0700)]
experimental C API: Fix compilation failure in Windows.

The functions added in
https://github.com/tensorflow/tensorflow/commit/be917027e37c5e8f21f6ba07f24bdbf072cf6dfd
are temporary, and their existence breaks compilation in MSVC because of
https://docs.microsoft.com/en-us/cpp/c-language/maximum-string-length
and
https://docs.microsoft.com/en-us/cpp/error-messages/compiler-errors-1/compiler-error-c2026

So just disabling it in Windows for now.

PiperOrigin-RevId: 192391164

6 years agoCheckpointable: remove colocation constraints from restore ops
Allen Lavoie [Wed, 11 Apr 2018 02:01:33 +0000 (19:01 -0700)]
Checkpointable: remove colocation constraints from restore ops

Mystery solved thanks to log_device_placement.

PiperOrigin-RevId: 192389574

6 years agoRemove BN workaround for resource variable gradients bug that was recently fixed.
Francois Chollet [Wed, 11 Apr 2018 01:50:58 +0000 (18:50 -0700)]
Remove BN workaround for resource variable gradients bug that was recently fixed.

PiperOrigin-RevId: 192388867

6 years agoMerge changes from github.
Michael Case [Wed, 11 Apr 2018 01:44:13 +0000 (18:44 -0700)]
Merge changes from github.

PiperOrigin-RevId: 192388250

6 years agoFix for users who were passing `Dimension` type as `units` arg in `Dense`.
Francois Chollet [Wed, 11 Apr 2018 01:41:37 +0000 (18:41 -0700)]
Fix for users who were passing `Dimension` type as `units` arg in `Dense`.

PiperOrigin-RevId: 192387984

6 years agoTFTS: De-flake the LSTM test
Allen Lavoie [Wed, 11 Apr 2018 01:04:20 +0000 (18:04 -0700)]
TFTS: De-flake the LSTM test

Disabling the value-based check for now. Hopefully the shapes are deterministic.

PiperOrigin-RevId: 192383553

6 years agoForward the status from LookupResource to GetInputTensorFromVariable rather than...
Allen Lavoie [Wed, 11 Apr 2018 00:57:19 +0000 (17:57 -0700)]
Forward the status from LookupResource to GetInputTensorFromVariable rather than returning a generic error status

PiperOrigin-RevId: 192382499

6 years ago[XLA] Redesign: implement Conj.
A. Unique TensorFlower [Wed, 11 Apr 2018 00:49:51 +0000 (17:49 -0700)]
[XLA] Redesign: implement Conj.

PiperOrigin-RevId: 192381481

6 years ago[XLA] Redesign: implement SliceInDim.
A. Unique TensorFlower [Wed, 11 Apr 2018 00:47:13 +0000 (17:47 -0700)]
[XLA] Redesign: implement SliceInDim.

PiperOrigin-RevId: 192381080

6 years ago[XLA] Redesign: implement ReduceAll.
A. Unique TensorFlower [Wed, 11 Apr 2018 00:44:48 +0000 (17:44 -0700)]
[XLA] Redesign: implement ReduceAll.

PiperOrigin-RevId: 192380688

6 years ago[XLA] Redesign: implement and test SelectAndScatter.
A. Unique TensorFlower [Wed, 11 Apr 2018 00:41:56 +0000 (17:41 -0700)]
[XLA] Redesign: implement and test SelectAndScatter.

PiperOrigin-RevId: 192380121

6 years agoUpdate LogToSTDErr for TF Lite usage
Austin Anderson [Wed, 11 Apr 2018 00:37:53 +0000 (17:37 -0700)]
Update LogToSTDErr for TF Lite usage

PiperOrigin-RevId: 192379483

6 years agoAdd types to error message in case of mismatch. NFC.
Jacques Pienaar [Tue, 10 Apr 2018 23:47:08 +0000 (16:47 -0700)]
Add types to error message in case of mismatch. NFC.

PiperOrigin-RevId: 192370979

6 years agoAdd a thread-safe producer-consumer queue.
Michael Kuperstein [Tue, 10 Apr 2018 23:45:19 +0000 (16:45 -0700)]
Add a thread-safe producer-consumer queue.

PiperOrigin-RevId: 192370670

6 years agoFix issue with gradients of resource variables in cond.
Alexandre Passos [Tue, 10 Apr 2018 23:35:58 +0000 (16:35 -0700)]
Fix issue with gradients of resource variables in cond.

PiperOrigin-RevId: 192369091

6 years ago[XLA] Redesign: implement and test ReduceWindow.
A. Unique TensorFlower [Tue, 10 Apr 2018 23:32:05 +0000 (16:32 -0700)]
[XLA] Redesign: implement and test ReduceWindow.

PiperOrigin-RevId: 192368401

6 years agoUpdate programmers guide
A. Unique TensorFlower [Tue, 10 Apr 2018 23:31:46 +0000 (16:31 -0700)]
Update programmers guide

PiperOrigin-RevId: 192368335

6 years agoSupporting FakeQuant num_bits and getting the fake quant op matching tensorflow.
A. Unique TensorFlower [Tue, 10 Apr 2018 23:26:05 +0000 (16:26 -0700)]
Supporting FakeQuant num_bits and getting the fake quant op matching tensorflow.

PiperOrigin-RevId: 192367307

6 years ago[XLA] GRPC service definition.
Chris Leary [Tue, 10 Apr 2018 23:12:19 +0000 (16:12 -0700)]
[XLA] GRPC service definition.

PiperOrigin-RevId: 192364932

6 years agoFix `nn` module RNN namespace issues.
Francois Chollet [Tue, 10 Apr 2018 23:11:38 +0000 (16:11 -0700)]
Fix `nn` module RNN namespace issues.

PiperOrigin-RevId: 192364808

6 years agoAllow passing allow_custom_ops for toco_convert.
Mingxing Tan [Tue, 10 Apr 2018 22:59:39 +0000 (15:59 -0700)]
Allow passing allow_custom_ops for toco_convert.

PiperOrigin-RevId: 192362688

6 years agoCleaning up _distributed_apply now the device policy is unnecessary
Alexandre Passos [Tue, 10 Apr 2018 22:49:03 +0000 (15:49 -0700)]
Cleaning up _distributed_apply now the device policy is unnecessary

PiperOrigin-RevId: 192360913

6 years agoIn `get_variable`, nest the choice to use `ResourceVariable` under an
Akshay Agrawal [Tue, 10 Apr 2018 22:48:15 +0000 (15:48 -0700)]
In `get_variable`, nest the choice to use `ResourceVariable` under an
`init_scope`.

This makes sure that, when executing eagerly, calls to `get_variable` in a
`defun`-compiled function retrieve `ResourceVariable`s instead of `Variables`.

PiperOrigin-RevId: 192360775

6 years agoSmall changes to testing code, plus a new binary to check diff from command line.
A. Unique TensorFlower [Tue, 10 Apr 2018 22:45:37 +0000 (15:45 -0700)]
Small changes to testing code, plus a new binary to check diff from command line.

PiperOrigin-RevId: 192360373

6 years ago[XLA] Redesign: implement and test Conv.
A. Unique TensorFlower [Tue, 10 Apr 2018 22:39:37 +0000 (15:39 -0700)]
[XLA] Redesign: implement and test Conv.

PiperOrigin-RevId: 192359226

6 years agoAllow negative feature values in computation for `sum` combiner.
A. Unique TensorFlower [Tue, 10 Apr 2018 22:23:05 +0000 (15:23 -0700)]
Allow negative feature values in computation for `sum` combiner.

PiperOrigin-RevId: 192355950

6 years agoAdd quantized LogSoftmax.
A. Unique TensorFlower [Tue, 10 Apr 2018 22:01:49 +0000 (15:01 -0700)]
Add quantized LogSoftmax.

PiperOrigin-RevId: 192352432

6 years agoRun EvaluateNodes for ModelPrunerTest_StopGradientPruning. Also updated the test...
A. Unique TensorFlower [Tue, 10 Apr 2018 21:51:54 +0000 (14:51 -0700)]
Run EvaluateNodes for ModelPrunerTest_StopGradientPruning. Also updated the test fixture to inherit from GrapplerTest.

PiperOrigin-RevId: 192350828

6 years agoParseOpData returns kTfLiteError when error happens.
Yu-Cheng Ling [Tue, 10 Apr 2018 21:24:51 +0000 (14:24 -0700)]
ParseOpData returns kTfLiteError when error happens.

PiperOrigin-RevId: 192346224

6 years agoBug fix for statistical_testing:
A. Unique TensorFlower [Tue, 10 Apr 2018 21:22:33 +0000 (14:22 -0700)]
Bug fix for statistical_testing:
  - Max/Min computations should be done over the sample dimension.
  - Change dominate check to be greater_equal instead of greater (for matching dimensions).

PiperOrigin-RevId: 192345809

6 years agoAdd missing import for RNNClassifier
A. Unique TensorFlower [Tue, 10 Apr 2018 21:16:36 +0000 (14:16 -0700)]
Add missing import for RNNClassifier

PiperOrigin-RevId: 192344760

6 years ago[XLA] Redesign: implement and test Rev, BitcastConvertType, Map, and ReducePrecision.
A. Unique TensorFlower [Tue, 10 Apr 2018 21:04:29 +0000 (14:04 -0700)]
[XLA] Redesign: implement and test Rev, BitcastConvertType, Map, and ReducePrecision.

PiperOrigin-RevId: 192342686

6 years agoPad support for quantized zero.
Suharsh Sivakumar [Tue, 10 Apr 2018 21:02:02 +0000 (14:02 -0700)]
Pad support for quantized zero.

PiperOrigin-RevId: 192342172

6 years agotest previously untested eval codepaths.
A. Unique TensorFlower [Tue, 10 Apr 2018 20:59:49 +0000 (13:59 -0700)]
test previously untested eval codepaths.

PiperOrigin-RevId: 192341561

6 years agoHandle duplicate features by coalescing them together into a single feature.
A. Unique TensorFlower [Tue, 10 Apr 2018 20:57:00 +0000 (13:57 -0700)]
Handle duplicate features by coalescing them together into a single feature.

PiperOrigin-RevId: 192341065

6 years agoRefactor layers:
Francois Chollet [Tue, 10 Apr 2018 20:49:37 +0000 (13:49 -0700)]
Refactor layers:
- tf.layers layers now subclasses tf.keras.layers layers.
- tf.keras.layers is now agnostic to variable scopes and global collections (future-proof). It also uses ResourceVariable everywhere by default.
- As a result tf.keras.layers is in general lower-complexity, with fewer hacks and workarounds. However some of current code is temporary (variable creation should be moved to Checkpointable, arguably, and there are some dependency issues that will require later refactors).
- The legacy tf.layers layers behavior is kept, with references to variable scopes and global collections injected in the subclassed tf.layers.base.Layer class (the content of tf.layers.base.Layer is the complexity differential between the old implementation and the new one).

Note: this refactor does slightly change the behavior of tf.layers.base.Layer, by disabling extreme edge-case behavior that either has long been invalid, or is dangerous and should most definitely be disabled. This will not affect any users since such behaviors only existed in the base Layer unit tests. The behaviors disabled are:
- Option to create reusable variables in `call` (already invalid for some time).
- Option to use a variable scope to create layer variables outside of the layer while not having the layer track such variables locally.
PiperOrigin-RevId: 192339798

6 years agoUpdate declaration order in staging ops Buffer class according to C++ style guide
Smit Hinsu [Tue, 10 Apr 2018 20:32:38 +0000 (13:32 -0700)]
Update declaration order in staging ops Buffer class according to C++ style guide

PiperOrigin-RevId: 192336966

6 years agoCheckpointable: wrap restore ops in init_scope
Allen Lavoie [Tue, 10 Apr 2018 19:54:03 +0000 (12:54 -0700)]
Checkpointable: wrap restore ops in init_scope

This should make restore() work with defun-wrapped code, when variables are
created inside the function. Just lifts the restore code into the outer
context. Adds a test for it.

PiperOrigin-RevId: 192331065

6 years agoEmploy array flat sizes more directly in reference_ops.
A. Unique TensorFlower [Tue, 10 Apr 2018 19:28:56 +0000 (12:28 -0700)]
Employ array flat sizes more directly in reference_ops.

PiperOrigin-RevId: 192327464

6 years agoUpgrade gRPC version and fix file duplication
Noah Eisen [Tue, 10 Apr 2018 19:28:04 +0000 (12:28 -0700)]
Upgrade gRPC version and fix file duplication

This bumps the gRPC version used in OSS Tensorflow to pick up grpc/grpc#14541, which exposes gRPC serialization classes which were previously hidden in an internal namespace. Using these files eliminates files duplicated from gRPC repo

PiperOrigin-RevId: 192327358

6 years agoExpand list of value-preserving ops. This will increase the number of graphs where...
A. Unique TensorFlower [Tue, 10 Apr 2018 19:05:06 +0000 (12:05 -0700)]
Expand list of value-preserving ops. This will increase the number of graphs where we can apply the involution and mul->conv fusion optimizations.

PiperOrigin-RevId: 192323712

6 years agoEnable loop-invariant node motion in the Grappler loop optimizer.
A. Unique TensorFlower [Tue, 10 Apr 2018 18:57:40 +0000 (11:57 -0700)]
Enable loop-invariant node motion in the Grappler loop optimizer.
Thanks to the team at Alibaba, who contributed the original version of this code.

PiperOrigin-RevId: 192322484

6 years agoFix bug in TFLite Interpreter python interface
A. Unique TensorFlower [Tue, 10 Apr 2018 18:33:45 +0000 (11:33 -0700)]
Fix bug in TFLite Interpreter python interface

PiperOrigin-RevId: 192318426

6 years ago[XLA] Fix the size of the data returned from Literal for sparse literals.
A. Unique TensorFlower [Tue, 10 Apr 2018 18:20:38 +0000 (11:20 -0700)]
[XLA] Fix the size of the data returned from Literal for sparse literals.

PiperOrigin-RevId: 192315888