Branch 141962539 by martinwicke · Pull Request #6296 · tensorflow/tensorflow

martinwicke · 2016-12-14T02:29:29Z

No description provided.

…ctor.is_constant_jacobian. Change: 141581989

Change: 141583614

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141589815

…42 because they cause test timeouts. Change: 141590858

…_top_k. Change: 141592580

…ible to do so correctly. Change: 141599116

Change: 141601347

Change: 141601412

Must compute dimension size explicitly rather than using [-1] in order to deal with all multidimensional Tensors with dimensions after first equal to zero. Change: 141601835

Change: 141605210

Change: 141611156

Change: 141611875

Change: 141612098

Change: 141613129

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141616518

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141619037

Without names they can't be refactored. Change: 141621006

Change: 141622018

op and the python use in the same change violates forwards compatibility. Keep the op registered. Change: 141622306

Change: 141625644

This should make bazel fetch significiantly faster and more reliable. Change: 141627448

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141630998

…n.static_rnn. Change: 141658438

Previously, is_inf_or_nan errors out if the dtype of the tensor (numpy array) is not float, complex or integer. This CL adds guard against such data types. Change: 141671477

…on in another partition. To order to work correctly with dedup of single-source-multiple-destination, we require that source node can only have one outgoing control edges. Change: 141675118

…es into a while loop. Change: 141679129

Making self._features_info vary by mode, as implemented, causes various test failures. We can consider doing it (more carefully) but need to revert for now as the broken build blocks other CLs. Change: 141702821

Change: 141703677

Change: 141704392

Change: 141721771

Change: 141934447

Change: 141935716

…ict. Allow initial state to be specified by returning state tensors from input_fn. Change: 141937736

…hat transform string features to IDs, where the mapping is defined by a vocabulary file. Change: 141945123

For some reason I thought the name was broken...so I wrote this test. Might as well use it. Change: 141945959

Usage examples: tfdbg> pt cross_entropy/Log:0 > /tmp/log_0.txt tfdbg> pt cross_entropy/Log:0[0, :] > /tmp/log_0_slice.txt tfdbg> li -r cross_entropy/Log:0 >/tmp/log_0_inputs.txt tfdbg> ni cross_entropy/Log:0>/tmp/log_0_node_info.txt tfdbg> lt> /tmp/tensor_list.txt Change: 141949497

This change introduces a new `LocalMaster` class, which enables distributed TensorFlow sessions to communicate directly with a master in the same process, without making a loopback RPC call. This improves the performance on some distributed session microbenchmarks: * The median step time for a minimal step decreases from 392us to 273us. * The median step time for a step that feeds 4MB of data decreases from 9.26ms to 5.26ms. This change is a first step towards addressing issue tensorflow#6256. The code contains TODO comments with several further potential improvements: * TODO(mrry): Add methods that avoid protobuf encoding the request/response objects where this affects performance. * TODO(mrry): Avoid closure creation/context switch overhead for synchronous invocation of Master methods. * TODO(mrry): Apply a similar change to the WorkerInterface. Change: 141950977

Change: 141951932

Before this CL, XLA GPU binaries can only be run with "blaze test". After this CL, they can be run directly via "blaze-bin/<executable>". Change: 141952101

…ct`. Change: 141961231

Change: 141962539

caisq · 2016-12-14T02:34:28Z

@tensorflow-jenkins test this please

caisq · 2016-12-14T03:56:40Z

//tensorflow/python/kernel_tests:variable_scope_test failure seems to be legit, reproducible failure. This failure is not seen on CL postsubmit tests. So I guess it is caused by a clash with a external PR. This one looks like a candidate: 0958bb7

martinwicke · 2016-12-14T04:59:15Z

I'm fixing it, slowly. I'll handle this, you should go sleep.

martinwicke · 2016-12-14T05:00:09Z

There was a large change affecting the behavior if zeros_initializer by @itsmeolivia, which the external PR wasn't aware of. Soon...

martinwicke · 2016-12-14T09:20:48Z

@mrry can you help with the cmake failure? I think it's new kernels in contrib, how should these be added to the cmake build?

mrry · 2016-12-14T16:33:44Z

You need to add a line here:

GENERATE_PYTHON_OP_LIB("contrib_tensor_forest_ops"
  DESTINATION ${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/contrib/tensor_forest/python/ops/gen_tensor_forest_ops.py)

It looks like the kernels are already built (see here), but the op registrations are missing, so we'll need another line here like:

GENERATE_CONTRIB_OP_LIBRARY(tensor_forest "${tensorflow_source_dir}/tensorflow/contrib/tensor_forest/ops/tensor_forest_ops.cc")

martinwicke · 2016-12-14T18:08:51Z

Thank you, that worked.

Joshua V. Dillon and others added 30 commits December 9, 2016 12:05

Support TransformedDistribution entropy in the limited case when bije…

a3f839e

…ctor.is_constant_jacobian. Change: 141581989

Update ops-related pbtxt files.

71b28db

Change: 141583614

Change references to TensorArray.pack, TensorArray.unpack to TensorAr…

66907af

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141589815

Remove some of the deprecations added to logging_ops.py in cl/1415159…

6d400d4

…42 because they cause test timeouts. Change: 141590858

Remove an unnecessary assert in metrics.streaming_sparse_precision_at…

c49e52c

…_top_k. Change: 141592580

Prevent ability to take second derivs of ctc_loss_op until it is poss…

760803a

…ible to do so correctly. Change: 141599116

Add missing test rule for draw_bounding_box_op

b48ec9a

Change: 141601347

Add missing test for adagrad_da

da9f662

Change: 141601412

boolean_mask BUGFIX.

996d477

Must compute dimension size explicitly rather than using [-1] in order to deal with all multidimensional Tensors with dimensions after first equal to zero. Change: 141601835

Update generated Python Op docs.

d499f52

Change: 141605210

Fix documentation link in SavedModel.

1da7d8a

Change: 141611156

Add missing test rule for registry_test

4cffce0

Change: 141611875

Automated rollback of change 140909396

e390dcd

Change: 141612098

Update Bazel version to 0.4.2

892dc07

Change: 141613129

Change references to TensorArray.pack, TensorArray.unpack to TensorAr…

6da1cfa

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141616518

Change references to TensorArray.pack, TensorArray.unpack to TensorAr…

35a58c8

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141619037

Name the nameless build rules

92cc148

Without names they can't be refactored. Change: 141621006

Update generated Python Op docs.

047072a

Change: 141622018

Undo the changes in python for the PreventGradient addition - adding the

f327050

op and the python use in the same change violates forwards compatibility. Keep the op registered. Change: 141622306

In sparse_to_dense_op.cc, include eigen before other libraries.

1fa6bc3

Change: 141625644

Use redundant download URLs

7bd9a1c

This should make bazel fetch significiantly faster and more reliable. Change: 141627448

Change references to TensorArray.pack, TensorArray.unpack to TensorAr…

d8f2a4b

…ray.stack and TensorArray.unstack since pack and unpack are getting deprecated. Also, I switched a few references to tf.pack/tf.unpack to tf.stack and tf.unstack. Change: 141630998

Update caller to move from tf.nn.rnn to (the identical) tf.contrib.rn…

13b0c97

…n.static_rnn. Change: 141658438

tfdbg: fix a dtype-related bug in debug_data.is_inf_or_nan

6c8fbf6

Previously, is_inf_or_nan errors out if the dtype of the tensor (numpy array) is not float, complex or integer. This CL adds guard against such data types. Change: 141671477

Dedup multiple control edges from one partition to a single destinati…

5d29db0

…on in another partition. To order to work correctly with dedup of single-source-multiple-destination, we require that source node can only have one outgoing control edges. Change: 141675118

Unconditionally invoke the function that removes external control edg…

6214c59

…es into a while loop. Change: 141679129

Revert breaking changes from tensorflow#5546

29c17b6

Making self._features_info vary by mode, as implemented, causes various test failures. We can consider doing it (more carefully) but need to revert for now as the broken build blocks other CLs. Change: 141702821

Switch tf.Learn to CheckpointV2 and fix tests.

86afb78

Change: 141703677

Update generated Python Op docs.

ff16c6e

Change: 141704392

tfdbg: add debug op DebugNumericSummary

ed814ca

Change: 141721771

mrry and others added 13 commits December 13, 2016 14:02

Fix use-after-free in the DirectSession timeout handling code.

2b771b9

Change: 141934447

Update comment to match current kernel registration macro.

df5d72d

Change: 141935716

Remove stale files.

1bd4c19

Decompose state into multiple tensors and return them in prediction d…

0b217fc

…ict. Allow initial state to be specified by returning state tensors from input_fn. Change: 141937736

Add sparse_column_with_vocabulary_file, to specify a feature column t…

f9e1a91

…hat transform string features to IDs, where the mapping is defined by a vocabulary file. Change: 141945123

LinearOperatorComposition: Add a test for the name.

fbc8107

For some reason I thought the name was broken...so I wrote this test. Might as well use it. Change: 141945959

Release notes update

f0a6d1e

Change: 141951932

Find libdevice files in .runfiles when TEST_SRCDIR is not set.

667a023

Before this CL, XLA GPU binaries can only be run with "blaze test". After this CL, they can be run directly via "blaze-bin/<executable>". Change: 141952101

Fix slowdown in graph building introduced by repeated calls to `inspe…

8d05819

…ct`. Change: 141961231

Internal only changes

2f0cd86

Change: 141962539

Merge BUILD file changes

527556c

googlebot added the cla: yes label Dec 14, 2016

Fix linter issues

c236f27

caisq approved these changes Dec 14, 2016

View reviewed changes

Fix initializer dtype problem

06ae1e2

martinwicke force-pushed the branch_141962539 branch from 9724654 to cd3cd23 Compare December 14, 2016 07:51

Fix initializer dtype problem

5a1498a

martinwicke force-pushed the branch_141962539 branch from cd3cd23 to 5a1498a Compare December 14, 2016 08:33

Add tensor_forest to cmake build

bc04100

martinwicke merged commit a304537 into tensorflow:master Dec 14, 2016

martinwicke deleted the branch_141962539 branch December 14, 2016 18:34

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Branch 141962539#6296

Branch 141962539#6296
martinwicke merged 112 commits into
tensorflow:masterfrom
martinwicke:branch_141962539

martinwicke commented Dec 14, 2016

Uh oh!

caisq commented Dec 14, 2016

Uh oh!

caisq commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

mrry commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants

Conversation

martinwicke commented Dec 14, 2016

Uh oh!

caisq commented Dec 14, 2016

Uh oh!

caisq commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

mrry commented Dec 14, 2016

Uh oh!

martinwicke commented Dec 14, 2016

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

20 participants