After
https://github.com/bazelbuild/bazel/commit/0b9ebfee8d7f3bd5cf68f4dba4790bf3fed9349d
//tensorflow/examples/android/... has to be excluded during bazel fetch.

Using the workaround mentioned in
https://github.com/bazelbuild/bazel/issues/2220, so that it works on
both Linux and Windows.

docs: add a troubleshooting section to the faq

Sometimes we could have an issue with a running docker, specifically with the TLS certificate.
It is important for a beginner to be aware of that.

Fixes #6181

* Make placement of constants follow consumers if they are all on the same device.

A generator node (for instant a constant) will now go to the same device
as its consumers, when those consumers are all on the same device.   Previously
a constant would only follow its consumer when there was only a single consumer.

* Update after review comments from vrv

* Further change after review comments.

* Update variable name.

* Fix up test to match the new fake CPU/GPU names

* sequence loss function for seq2seq loss

* loss raises and docstring updated

* detailed docstring

* moving initializer one line down

* change order of arguments, sparse_crossent

* Use tf.stack instead of tf.pack.

* Have `write_graph` return the output path of file

Since the function is performing `os.path.join()` for the user, it makes
sense to return the location of the final file, or else the user will
have to call `os.path.join()` again on their own.

* Adjust WriteGraphTest to check returned path

* Assert path exists; "/".join -> os.path.join

Theoretically more Windows compatible

* Disable building the MatrixDiag and OneHot kernels for GPU on Windows.

This is a temporary workaround for issue #6509, while the root cause
of the runtime errors is under investigation.

* Handle not-found target properties.

* Fix typo that caused everything to be built twice.

* Re-disable MatrixDiag test for now.

tensorflow/contrib/makefile/build_all_android.sh disagrees about the environment variable to use. This seems the easiest fix with the least chance of breaking anything else.

* Update einsum to check whether uncompacted_shape has multiple None values

* Add more tests

Signed-off-by: NNorman Heckscher <norman.heckscher@gmail.com>

Closes #5611

When read_result is TRUE, result should be set to bytes_read.

Branch 144396000

Conflict tensorflow/contrib/layers/python/layers/layers.py:
Preserved both indentation on "decay" change and doc fix on "center".

Change: 144396000

Change: 144395527

Change: 144395110

Change: 144394762

This commit creates a benchmark for the resize_bicubic op on the example sizes used in the inception_train script. The current performance on my desktop is:
 - bicubic_749_603: 4.98ms/image
 - bicubic_141_186: 4.79ms/image
 - bicubic_183_229: 4.80ms/image
Change: 144392656

Change: 144392019

Temporary fix of bazel dependencies when including TensorFlow as a submodule. Undo change when this use-case is supported in native.http_archive in Bazel.
Change: 144390772

Useful for simplifying a number of flatten -> map -> recreate structure calls.
Change: 144388829

Change: 144387537

Change: 144384783

Change: 144384102

Change: 144384086

Add support for marking Xla computations as stateful.
Add a store for xla::ChannelHandles in XlaCompiler.
Don't mark _Send/_Recv for XLA computation.
Change: 144382814

- Handle more functions:
  tf.svd
  tf.batch_matmul
  tf.nn.softmax_cross_entropy_with_logits,
  tf.nn.sparse_softmax_cross_entropy_with_logits,
  tf.nn.sigmoid_cross_entropy_with_logits": [
- Handle in-place file modification correctly (and add test).
- Handle raw attribute lookups i.e. lists of functions
  `foo = [tf.mul]`  can be upgraded to `foo = [tf.multiply]`
Change: 144381716

[TF:XLA] Lower severity of warning if an XLA_GPU device cannot be created to reduce log spam in opensource tests.
Change: 144381041

Change: 144378661

Change: 144376783

Change: 144375998

Change: 144371549

Reformatted core/platform/posix/port.cc as a test.
Change: 144368934