Yong Tang [Tue, 17 Apr 2018 23:44:53 +0000 (16:44 -0700)]
Replace raw_input/input with six.moves.input for python 2/3 (#18461)
In python 3 input is the equivalent of raw_input in python 2.
This fix is an enhancement to replace raw_input/input with
six.moves.input, which is compatible with both python 2 and python 3.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 23:44:21 +0000 (16:44 -0700)]
Fix shape validation error with tf.nn.conv3d_transpose (#18465)
* Fix shape validation error with tf.nn.conv3d_transpose
This fix tries to address the issue raised in 18460.
In `tf.nn.conv3d_transpose` when list or np array is passed,
the validate of the output shape with filter shape uses
`output_shape[4]` (channel). This will not work with
`data_format='NCDHW'`.
This fix fixes the issue by replace `output_shape[4]` with `output_shape[axis]`.
This fix also adds a test case. Before this fix, the test case will fail.
This fix fixes 18460.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
* Add test case for output and filter shape check in conv3d_transpose
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
* Fix pylint issue
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
* Also fix the error message
shengfuintel [Tue, 17 Apr 2018 23:44:10 +0000 (16:44 -0700)]
Fixed the bug in mkl_input_conversion_op when reorder is not needed (#18498)
AG Ramesh [Tue, 17 Apr 2018 23:43:53 +0000 (16:43 -0700)]
[INTEL MK] Updating MKL CPU CI build and test. (#18513)
* Setting KMP_BLOCKTIME to 0
* Adding comments
Martin Wicke [Tue, 17 Apr 2018 23:26:41 +0000 (16:26 -0700)]
Merge pull request #18618 from yongtang/
04172018-duplicate-random_ops
Remove duplicate imports in all python source
Derek Murray [Tue, 17 Apr 2018 21:59:58 +0000 (14:59 -0700)]
[tf.data] Fix a device placement issue in `prefetch_to_device()`. (#18607)
* [tf.data] Fix a device placement issue in `prefetch_to_device()`.
Previously, the `iterator_get_device()` op was being infeasibly colocated with
both the iterator and placed on the prefetch target device. Move the
construction of that op outside the `with device():` block to fix this.
Also enable the relevant test to run as a CUDA test.
* Import the cuda_py_test rule.
Martin Wicke [Tue, 17 Apr 2018 21:59:50 +0000 (14:59 -0700)]
Merge pull request #18615 from yongtang/
04162018-TFRecordDataset-shape
Add shape check to TFRecordDataset
Yifei Feng [Tue, 17 Apr 2018 21:54:02 +0000 (14:54 -0700)]
Merge pull request #18613 from yifeif/branch_193234819
Branch
193234819
Yong Tang [Tue, 17 Apr 2018 21:36:21 +0000 (21:36 +0000)]
Clean up remaining issues.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 21:36:05 +0000 (21:36 +0000)]
Fix duplicate import in kmeans_test.py
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 21:33:27 +0000 (21:33 +0000)]
Remove duplicate imports in several places
Wrote a script to scan throught the python files in the repo,
and found the remaining duplicate imports in some python files like:
```
from tensorflow.python.ops import random_ops
-from tensorflow.python.ops import random_ops
from tensorflow.python.util.deprecation import deprecated
```
This fix removed all of them for duplicate imports.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Akshay Modi [Tue, 17 Apr 2018 21:40:32 +0000 (14:40 -0700)]
Merge pull request #18504 from akshaym/eager-custom-ops
Use eager compatible wrappers in load_library for custom ops
Martin Wicke [Tue, 17 Apr 2018 21:14:41 +0000 (14:14 -0700)]
Merge pull request #18614 from yongtang/
04172018-duplicate-import
Remove duplicate import in compat.py
Martin Wicke [Tue, 17 Apr 2018 20:34:33 +0000 (13:34 -0700)]
Merge pull request #18561 from yongtang/18529-doc-cast
Update docs for tf.cast with supported types
Martin Wicke [Tue, 17 Apr 2018 20:33:05 +0000 (13:33 -0700)]
Merge pull request #18563 from yongtang/
04162018-gemmlowp
Update gemmlowp version for cmake build
Martin Wicke [Tue, 17 Apr 2018 20:18:06 +0000 (13:18 -0700)]
Merge pull request #17562 from giuscri/typo
Fix markdown error in layers tutorial.
Martin Wicke [Tue, 17 Apr 2018 20:17:41 +0000 (13:17 -0700)]
Merge pull request #17512 from imsheridan/fix_broken_link_graph
Fix broken link of distributed in graphs.md
Yifei Feng [Tue, 17 Apr 2018 20:05:09 +0000 (13:05 -0700)]
Update clip_ops_test.py
annarev [Tue, 17 Apr 2018 19:54:30 +0000 (12:54 -0700)]
Merge pull request #17395 from facaiy/ENH/adamax_optimizer
add AdaMax optimizer
Yong Tang [Tue, 17 Apr 2018 19:48:42 +0000 (19:48 +0000)]
Duplicate imports in histogram_ops.py
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 19:47:25 +0000 (19:47 +0000)]
Remove duplicate import in compat.py
Noticed there are a couple of places in compat.py that
have duplicate import:
```
from tensorflow.python.util.tf_export import tf_export
from tensorflow.python.util.tf_export import tf_export
```
This fix remove duplicate imports.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yifei Feng [Tue, 17 Apr 2018 19:18:44 +0000 (12:18 -0700)]
Merge commit for internal changes
Tang, Wenyi [Tue, 17 Apr 2018 19:17:48 +0000 (03:17 +0800)]
Complement cmake script to compile tensorflow with mkl and mkldnn on Windows (#16936)
* Add build batch for windows
* Automaticaly find python, cuda, mkl runtimes in PATH
* auto select cmake generator
* Add external library mkldnn. Add options for mkl and mkldnn
* fix syntax error in make.bat
* Fix errorlevel syntex bug in make.bat
* Add /arch:avx2 flags to enable avx2 on windows
* Revert to keep `tensprflow_WIN_CPU_SIMD_OPTIONS` unchanged, add an option `tensorflow_ENABLE_MKL_SUPPORT` to include MKL compilation. Still specify SIMD flags by setting `tensorflow_WIN_CPU_SIMD_OPTIONS` to such as '/arch:AVX2'
* Fix a mistake of CUDA path in make.bat
* resolve conflict in mkl_cpu_allocator.h
* Improve error detection
* Use where /Q to detect cmd environment
* fix "ELSE IF" the syntax error in make.bat
* update README.md, wrap windows based codes by #ifdef _WIN32
* unistd.h is not needed in mkl_cpu_allocator.h any more in master branch
* Remove inline of kMaxLimitStr, which causes compile error in VS2015
* Add static_cast in to fix compile error
* remove make.bat
* Removed make.bat description part
Yifei Feng [Tue, 17 Apr 2018 19:07:33 +0000 (12:07 -0700)]
Automated g4 rollback of changelist
193168327
PiperOrigin-RevId:
193234819
Pavithra Vijay [Tue, 17 Apr 2018 19:06:50 +0000 (12:06 -0700)]
Moving gradient registration for CudnnRNN op from contrib to core.
PiperOrigin-RevId:
193234663
A. Unique TensorFlower [Tue, 17 Apr 2018 18:53:29 +0000 (11:53 -0700)]
Fixes and cleanup to support more complex quantized models and adds PropagateFakeQuantNumBits.
PiperOrigin-RevId:
193232630
Martin Wicke [Tue, 17 Apr 2018 18:54:48 +0000 (11:54 -0700)]
Merge pull request #18601 from yongtang/18598-tf.compat.as_str
Fix tf.compat.as_str returns bytes issue in Python 3
Martin Wicke [Tue, 17 Apr 2018 18:52:39 +0000 (11:52 -0700)]
Merge pull request #18581 from ManHyuk/fix_typo
Fix typo
Mingxing Tan [Tue, 17 Apr 2018 18:48:43 +0000 (11:48 -0700)]
Replace decode_image with decode_jpeg to avoid ValueError in datasets programmers guide.
PiperOrigin-RevId:
193231717
A. Unique TensorFlower [Tue, 17 Apr 2018 18:05:57 +0000 (11:05 -0700)]
Fixes a comment in tf.contrib.seq2seq.monotonic_attention().
PiperOrigin-RevId:
193224285
Yifei Feng [Tue, 17 Apr 2018 17:53:07 +0000 (10:53 -0700)]
Enable consumption of GIT_TAG_OVERRIDE env var in release build script. (#18579)
Enable consumption of GIT_TAG_OVERRIDE env var in release build script.
Akshay Modi [Tue, 17 Apr 2018 17:32:47 +0000 (10:32 -0700)]
Avoid ToString() in Eager's TFE_Execute.
Also use InlinedVector instead of std::vector for non-async path
Before:
Benchmark Time(ns) CPU(ns) Iterations
-------------------------------------------------------------
BM_Execute/0 1895 1898 360200 Execute
BM_Execute/1 1193 1942 358322 ExecuteAsync
BM_ExecuteFunction/0 5812 5825 100000 ExecuteFunction
BM_ExecuteFunction/1 5015 5374 100000 ExecuteFunctionAsync
After:
Benchmark Time(ns) CPU(ns) Iterations
-------------------------------------------------------------
BM_Execute/0 1604 1607 428262 Execute
BM_Execute/1 1150 1765 404821 ExecuteAsync
BM_ExecuteFunction/0 5615 5626 100000 ExecuteFunction
BM_ExecuteFunction/1 5111 5476 100000 ExecuteFunctionAsync
PiperOrigin-RevId:
193218331
A. Unique TensorFlower [Tue, 17 Apr 2018 17:13:09 +0000 (10:13 -0700)]
Change the contract of dynamic_builtin to reject all functions it can't process.
PiperOrigin-RevId:
193215246
Yong Tang [Tue, 17 Apr 2018 16:15:20 +0000 (16:15 +0000)]
Pylint fix
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Max Galkin [Tue, 17 Apr 2018 16:19:06 +0000 (09:19 -0700)]
Estimate IdentityN as Identity.
PiperOrigin-RevId:
193207469
Yong Tang [Tue, 17 Apr 2018 16:14:50 +0000 (16:14 +0000)]
Removed unnneded tf_export
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 16:10:27 +0000 (16:10 +0000)]
Fix tf.compat.as_str returns bytes issue in Python 3
This fix tries to address the issue raised in 18598 where
tf.compat.as_str returns bytes (vs. str) in Python 3.
The issue was that `tf_export` decorator:
```
@tf_export('compat.as_bytes', 'compat.as_str')
```
could not be assigned to `as_bytes` or `as_text`
based on python 2 or 3.
This fix invokes tf_export explicitly based on `_six.PY2`
(for python 2/3) so that `as_str` calls `as_bytes` or `as_text`
conditionally.
This fix fixes 18598.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
A. Unique TensorFlower [Tue, 17 Apr 2018 16:02:14 +0000 (09:02 -0700)]
Fix incorrect rejection of xrange.
PiperOrigin-RevId:
193205016
A. Unique TensorFlower [Tue, 17 Apr 2018 14:41:09 +0000 (07:41 -0700)]
Add uint32/uint64 support to Gather op.
PiperOrigin-RevId:
193195939
A. Unique TensorFlower [Tue, 17 Apr 2018 14:07:16 +0000 (07:07 -0700)]
Fix the type info analysis to correctly process loops. Simplify the implementation by reusing some of the transformer base functions. Allow set_element_type to use literals. Add additional tests.
PiperOrigin-RevId:
193192409
A. Unique TensorFlower [Tue, 17 Apr 2018 09:23:10 +0000 (02:23 -0700)]
Automated g4 rollback of changelist
192842670
PiperOrigin-RevId:
193168327
Derek Murray [Tue, 17 Apr 2018 06:07:24 +0000 (23:07 -0700)]
[tf.data] Add an API for randomly sampling from multiple datasets.
Fixes #15999.
PiperOrigin-RevId:
193152683
Saurabh Saxena [Tue, 17 Apr 2018 03:28:02 +0000 (20:28 -0700)]
Make ParallelInterleaveDataset saveable.
PiperOrigin-RevId:
193142302
A. Unique TensorFlower [Tue, 17 Apr 2018 02:51:13 +0000 (19:51 -0700)]
Supply a dtype to super constructor, without which build() seems to crash.
PiperOrigin-RevId:
193139585
Younghee Kwon [Tue, 17 Apr 2018 02:10:10 +0000 (19:10 -0700)]
BoostedTreesEstimator in contrib: train_in_memory works with input_fns returning data.Dataset.
Only one batch of data is expected, so dataset.batch() is disallowed,
and dataset.repeat() will be ignored (only the first one would be used)
PiperOrigin-RevId:
193137094
manhyuk [Tue, 17 Apr 2018 01:52:16 +0000 (10:52 +0900)]
fix typo
manhyuk [Tue, 17 Apr 2018 01:48:52 +0000 (10:48 +0900)]
fix typo
manhyuk [Tue, 17 Apr 2018 01:46:37 +0000 (10:46 +0900)]
fix typo
Yifei Feng [Tue, 17 Apr 2018 01:41:28 +0000 (18:41 -0700)]
Remove proto imports in header files for core/kernels/hexagon.
The goal is to make kernels mostly independent of proto headers, which will let us lock down our .so imports.
PiperOrigin-RevId:
193134710
Shashi Shekhar [Tue, 17 Apr 2018 01:31:22 +0000 (18:31 -0700)]
Add a simple Profiler and instrument operator invocations in Interpreter.
PiperOrigin-RevId:
193133955
Yong Tang [Tue, 17 Apr 2018 01:32:41 +0000 (01:32 +0000)]
Sanitize with clan-format -i --style=Google
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 01:31:54 +0000 (01:31 +0000)]
Add shape check for buffer_size with TFRecordDataset
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 01:30:42 +0000 (01:30 +0000)]
Add shape check for compression_type in TFrecordDataset
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Tue, 17 Apr 2018 01:28:30 +0000 (01:28 +0000)]
Add shape check to TFRecordDataset
The inputs of TFRecordDataset have the requirements for shapes.
However, the check was not done in the shape function. This fix
adds shape checks whenever possible.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
A. Unique TensorFlower [Tue, 17 Apr 2018 01:03:05 +0000 (18:03 -0700)]
Adding min node weight regularization
PiperOrigin-RevId:
193131300
Michael Case [Tue, 17 Apr 2018 00:34:01 +0000 (17:34 -0700)]
Remove deprecated/unused python related Bazel options.
Since py_runtime was introduced, Bazel ignores options such as
--force_python2 and --python2_path. Deleting to clean stuff up and
make sure people are not misled.
PiperOrigin-RevId:
193127681
Derek Murray [Tue, 17 Apr 2018 00:25:12 +0000 (17:25 -0700)]
[tf.data] Sort the results of `tf.matching_files()` to enable `Dataset.list_files()` to be determinstic.
PiperOrigin-RevId:
193126572
Benoit Steiner [Tue, 17 Apr 2018 00:12:05 +0000 (17:12 -0700)]
Don't rely on graph contruction for an initial shape inference.
PiperOrigin-RevId:
193124836
A. Unique TensorFlower [Tue, 17 Apr 2018 00:05:18 +0000 (17:05 -0700)]
Add LinearOperatorKronecker, representing the Kronecker product.
PiperOrigin-RevId:
193123894
Martin Wicke [Tue, 17 Apr 2018 00:00:03 +0000 (17:00 -0700)]
Merge pull request #18575 from miaout17/fix-fft
Fix TFLite Makefile FFT2D dependency.
Anna R [Mon, 16 Apr 2018 23:41:12 +0000 (16:41 -0700)]
Internal change.
PiperOrigin-RevId:
193119953
A. Unique TensorFlower [Mon, 16 Apr 2018 23:34:18 +0000 (16:34 -0700)]
Port the list append into the operators module. Not enabled yet.
PiperOrigin-RevId:
193118940
Martin Wicke [Mon, 16 Apr 2018 23:30:50 +0000 (16:30 -0700)]
Merge pull request #17443 from yongtang/16263-softmax_cross_entropy-doc
Update the documentation of `softmax_cross_entropy`
Olivia Nordquist [Mon, 16 Apr 2018 23:24:19 +0000 (16:24 -0700)]
disabling flaky asan test
PiperOrigin-RevId:
193117611
A. Unique TensorFlower [Mon, 16 Apr 2018 23:22:03 +0000 (16:22 -0700)]
Internal change
PiperOrigin-RevId:
193117142
A. Unique TensorFlower [Mon, 16 Apr 2018 23:15:19 +0000 (16:15 -0700)]
Create copy of locals() before copying, since modifying locals does not always affect the values. https://docs.python.org/2/library/functions.html#locals.
PiperOrigin-RevId:
193116254
A. Unique TensorFlower [Mon, 16 Apr 2018 23:04:43 +0000 (16:04 -0700)]
Replacing the current inner Cholesky decomposition loop with a While loop rolled version.
This will allow for much larger Cholesky decompositions (and thus matrix inversions) than previously possible on TPU because of the use of rolled While loops so XLA compilation will no longer timeout.
While there is a minor runtime performance decrease (now 25ms vs 15ms for a 500x500 matrix) the compilation time is significantly faster (12.8s vs 55.2s for a 500x500 matrix.)
PiperOrigin-RevId:
193114816
Mark Daoust [Mon, 16 Apr 2018 22:56:41 +0000 (15:56 -0700)]
Merge branch 'master' into typo
A. Unique TensorFlower [Mon, 16 Apr 2018 22:52:54 +0000 (15:52 -0700)]
Fix trace collection to properly remove the suffix.
PiperOrigin-RevId:
193113074
Mark Daoust [Mon, 16 Apr 2018 22:53:08 +0000 (15:53 -0700)]
Curly-brace id's are inconsistently supported.
Curly-brace id's are inconsistently supported.
linking to the id of an html tag seems to be supported everywhere.
A. Unique TensorFlower [Mon, 16 Apr 2018 22:46:42 +0000 (15:46 -0700)]
Internal change.
PiperOrigin-RevId:
193112205
Yu-Cheng Ling [Mon, 16 Apr 2018 22:34:50 +0000 (15:34 -0700)]
Fix TFLite Makefile FFT2D dependency.
FFT2D dependency was introduced a while ago so Makefile no longer works
until this fix.
James Qin [Mon, 16 Apr 2018 21:52:41 +0000 (14:52 -0700)]
Increase softmax gpu unittest numeric stability
PiperOrigin-RevId:
193103363
Jiri Simsa [Mon, 16 Apr 2018 21:47:31 +0000 (14:47 -0700)]
Porting tests for `rpc_op` to OS.
PiperOrigin-RevId:
193102564
A. Unique TensorFlower [Mon, 16 Apr 2018 21:41:21 +0000 (14:41 -0700)]
Exposes InputPipelineConfig as tf.contrib.tpu.InputPipelineConfig.
This type is expected by the `per_host_input_for_training` argument of the TPUConfig constructor, but is not currently visible.
PiperOrigin-RevId:
193101540
Yong Tang [Mon, 16 Apr 2018 21:34:38 +0000 (21:34 +0000)]
Fix lite and makefile issue
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Martin Wicke [Mon, 16 Apr 2018 21:28:00 +0000 (14:28 -0700)]
Merge pull request #18558 from imsheridan/fix_argmax_dimension
Fix tf.argmax warnings on dimension argument by using axis instead
Martin Wicke [Mon, 16 Apr 2018 21:27:48 +0000 (14:27 -0700)]
Merge pull request #18547 from imsheridan/fix_deprecated_softmax
Fix the doc strings of nn.sampled_softmax_loss since it was deprecated
Martin Wicke [Mon, 16 Apr 2018 21:25:16 +0000 (14:25 -0700)]
Merge pull request #18570 from yongtang/18106-Bahdanau-normalized-dtype
Fix the issue with Bahdanau attention when normalized=True and dtype = float16/32
Olivia Nordquist [Mon, 16 Apr 2018 21:17:02 +0000 (14:17 -0700)]
disabling test that fails tensorflow.asan
PiperOrigin-RevId:
193097794
Yong Tang [Mon, 16 Apr 2018 20:16:06 +0000 (20:16 +0000)]
Update bazel
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
A. Unique TensorFlower [Mon, 16 Apr 2018 21:13:52 +0000 (14:13 -0700)]
Use fixed sized tensor arrays and max loop iterations in dynamic_decode if the user supplies it and if the inputs were created in an XLA context.
PiperOrigin-RevId:
193097293
A. Unique TensorFlower [Mon, 16 Apr 2018 20:54:03 +0000 (13:54 -0700)]
Simplify the recursion when processing unpackings.
PiperOrigin-RevId:
193094078
Martin Wicke [Mon, 16 Apr 2018 20:48:33 +0000 (13:48 -0700)]
Merge pull request #18562 from yongtang/
04152018-png
Update libpng to v1.6.34 for cmake build
Nick Desaulniers [Mon, 16 Apr 2018 20:36:55 +0000 (13:36 -0700)]
[XLA] Document and enforce reduction order of init value
All existing backends apply the init_value on the lhs, except for the evaluator. This causes problems for reductions which apply an identity function to a reduce or reduce window.
PiperOrigin-RevId:
193091323
A. Unique TensorFlower [Mon, 16 Apr 2018 20:32:12 +0000 (13:32 -0700)]
Adding several utility functions to TF2XLA to help with the Cholesky refactor. Mainly responsible for handling batching properly.
PiperOrigin-RevId:
193090634
A. Unique TensorFlower [Mon, 16 Apr 2018 20:22:58 +0000 (13:22 -0700)]
Update ops-related pbtxt files.
PiperOrigin-RevId:
193089301
Yong Tang [Mon, 16 Apr 2018 20:02:46 +0000 (20:02 +0000)]
Add gemmlowp to whitelist
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Mon, 16 Apr 2018 19:40:43 +0000 (19:40 +0000)]
Temporary comment out mirror.bazel.build for gemmlowp
Will reenable once the mirror is propagated.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Mon, 16 Apr 2018 19:25:05 +0000 (19:25 +0000)]
Update gemmlowp on bazel to sync cmake changes
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Mon, 16 Apr 2018 16:27:50 +0000 (16:27 +0000)]
Update gemmlowp version for cmake build
The gemmlowp has been updated in bazel, though
cmake version was not updated. This fix updates
gemmlowp in cmake so that cmake and bazel versions
are synced.
The fix has been tested on Linux:
```
tensorflow/tools/ci_build/ci_build.sh CMAKE tensorflow/tools/ci_build/builds/cmake.sh
```
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
A. Unique TensorFlower [Mon, 16 Apr 2018 20:04:23 +0000 (13:04 -0700)]
Simplify the implementation of break_statements.py
PiperOrigin-RevId:
193086371
Yong Tang [Sun, 1 Apr 2018 02:08:53 +0000 (02:08 +0000)]
Fix pylint issue
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Sun, 1 Apr 2018 02:07:30 +0000 (02:07 +0000)]
Add test case for Bahdanau attention when normalized=True and dtype = float16/32
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Sun, 1 Apr 2018 02:01:47 +0000 (02:01 +0000)]
Fix the issue with Bahdanau attention when normalized=True and dtype = float16/32
While revisiting 18016 I noticed that Bahdanau attention has a similiar
dtype mismatch issue when normalized=True. The issue comes from:
```
g = variable_scope.get_variable(
"attention_g", dtype=dtype,
initializer=math.sqrt((1. / num_units)))
```
where the initializer value does not work well with differnt dtype.
This fix converts changes the initializer to `init_ops.constant_initializer`
to address the issue, and adds additional test cases for it.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Younghee Kwon [Mon, 16 Apr 2018 19:56:14 +0000 (12:56 -0700)]
boosted_trees: Make some regularizer/hyper-params as inputs instead of attributes.
PiperOrigin-RevId:
193085059
Yong Tang [Sat, 14 Apr 2018 19:05:28 +0000 (19:05 +0000)]
Add missing header pnglibconf.h
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Sat, 14 Apr 2018 19:04:52 +0000 (19:04 +0000)]
Update library file names `libpng12` -> `libpng16`
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Yong Tang [Sat, 14 Apr 2018 19:00:36 +0000 (19:00 +0000)]
Update libpng to v1.6.34 for cmake
The libpng has been updated from v1.2.53 to v1.6.34 in PR 18299.
However, the cmake version of libpng has not been updated yet.
This fix updates the libpng for cmake to v1.6.34.
The fix is tested with cmake on linux:
```
tensorflow/tools/ci_build/ci_build.sh CMAKE tensorflow/tools/ci_build/builds/cmake.sh
```
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
Igor Saprykin [Mon, 16 Apr 2018 19:21:15 +0000 (12:21 -0700)]
Early TPU distribution strategy and the associated testing infrastructure.
PiperOrigin-RevId:
193080098