experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge changes from github.

Browse Source 
Additionally:
- change single quotes to double quotes to make path rewriting easier
- guard windows lib reference with PLATFORM_WINDOWS
- fixed failing kmeans test
Change: 141515942
This commit is contained in:
Andrew Harp 2016-12-08 20:05:49 -08:00 committed by TensorFlower Gardener
parent 90b72f4b2f
commit 1cb96893a6
256 changed files with 6163 additions and 2346 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
README.md
View File

@ -3,9 +3,9 @@
</div>
-----------------
| **`Linux CPU`** | **`Linux GPU PIP`** | **`Mac OS CPU`** | **`Android`** |
|-------------------|----------------------|------------------|----------------|
| [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-cpu)](https://ci.tensorflow.org/job/tensorflow-master-cpu) | [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-gpu_pip)](https://ci.tensorflow.org/job/tensorflow-master-gpu_pip) | [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-mac)](https://ci.tensorflow.org/job/tensorflow-master-mac) | [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-android)](https://ci.tensorflow.org/job/tensorflow-master-android) |
| **`Linux CPU`** | **`Linux GPU`** | **`Mac OS CPU`** | **`Windows CPU`** | **`Android`** |
|-----------------|---------------------|------------------|-------------------|---------------|
| [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-cpu)](https://ci.tensorflow.org/job/tensorflow-master-cpu) | [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-linux-gpu)](https://ci.tensorflow.org/job/tensorflow-master-linux-gpu) | [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-mac)](https://ci.tensorflow.org/job/tensorflow-master-mac) | [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-win-cmake-py)](https://ci.tensorflow.org/job/tensorflow-master-win-cmake-py) | [![Build Status](https://ci.tensorflow.org/buildStatus/icon?job=tensorflow-master-android)](https://ci.tensorflow.org/job/tensorflow-master-android) |
**TensorFlow** is an open source software library for numerical computation using
data flow graphs. Nodes in the graph represent mathematical operations, while
@ -33,10 +33,10 @@ and discussion.**
People who are a little more adventurous can also try our nightly binaries:
* Linux CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.11.0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.11.0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.11.0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/))
* Linux GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.11.0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.11.0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.11.0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/))
* Mac CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.11.0-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.11.0-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/))
* Mac GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.11.0-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.11.0-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/))
* Linux CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/))
* Linux GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/))
* Mac CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc0-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-0.12.0rc0-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/))
* Mac GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc0-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-mac/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-0.12.0rc0-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-mac-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-mac/))
* [Android](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-android/TF_BUILD_CONTAINER_TYPE=ANDROID,TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=NO_PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=android-slave/lastSuccessfulBuild/artifact/bazel-out/local_linux/bin/tensorflow/examples/android/tensorflow_demo.apk) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-android/TF_BUILD_CONTAINER_TYPE=ANDROID,TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=NO_PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=android-slave/))
#### *Try your first TensorFlow program*

108
RELEASE.md
View File

@ -6,9 +6,49 @@
semantics. tf.div is renamed to tf.division. New operators tf.truncatediv and
tf.truncatemod are available for achieving the previous C++ (truncation)
division/modulus semantics.
# Release 0.12.0
## Major Features and Improvements
* TensorFlow now builds and runs on Microsoft Windows (tested on Windows 10,
Windows 7, and Windows Server 2016). Supported languages include Python (via a
pip package) and C++. CUDA 8.0 and cuDNN 5.1 are supported for GPU
acceleration. Known limitations include: It is not currently possible to load
a custom op library. The GCS and HDFS file systems are not currently
supported. The following ops are not currently implemented:
DepthwiseConv2dNative, DepthwiseConv2dNativeBackpropFilter,
DepthwiseConv2dNativeBackpropInput, Dequantize, Digamma, Erf, Erfc, Igamma,
Igammac, Lgamma, Polygamma, QuantizeAndDequantize, QuantizedAvgPool,
QuantizedBatchNomWithGlobalNormalization, QuantizedBiasAdd, QuantizedConcat,
QuantizedConv2D, QuantizedMatmul, QuantizedMaxPool,
QuantizeDownAndShrinkRange, QuantizedRelu, QuantizedRelu6, QuantizedReshape,
QuantizeV2, RequantizationRange, and Requantize.
* Go: Experimental API in Go to create and execute graphs
(https://godoc.org/github.com/tensorflow/tensorflow/tensorflow/go)
* New checkpoint format becomes the default in `tf.train.Saver`. Old V1
checkpoints continue to be readable; controlled by the `write_version`
argument, `tf.train.Saver` now by default writes out in the new V2
format. It significantly reduces the peak memory required and latency
incurred during restore.
* Added a new library for library of matrix-free (iterative) solvers for linear
equations, linear least-squares, eigenvalues and singular values in
tensorflow/contrib/solvers. Initial version has lanczos bidiagonalization,
conjugate gradients and CGLS.
* Added gradients for `matrix_solve_ls` and `self_adjoint_eig`.
* Large cleanup to add second order gradient for ops with C++ gradients and
improve existing gradients such that most ops can now be differentiated
multiple times.
* Added a solver for ordinary differential equations,
`tf.contrib.integrate.odeint`.
* New contrib module for tensors with named axes, `tf.contrib.labeled_tensor`.
* Visualization of embeddings in TensorBoard.
## Breaking Changes to the API
* `BusAdjacency` enum replaced with a protocol buffer `DeviceLocality`. PCI bus
indexing now starts from 1 instead of 0, and bus_id==0 is used where previously
BUS_ANY was used.
indexing now starts from 1 instead of 0, and bus_id==0 is used where
previously BUS_ANY was used.
* `Env::FileExists` and `FileSystem::FileExists` now return a tensorflow::Status
intead of a bool. Any callers to this function can be converted to a bool
by adding .ok() to the call.
@ -16,8 +56,9 @@ BUS_ANY was used.
indicating its preferred use in language bindings for TensorFlow.
What was previously `TF_Session` has been renamed to `TF_DeprecatedSession`.
* Renamed TF_Port to TF_Output in the C API.
* Removes RegisterShape from public API. Use C++ shape function registration
instead.
* Removes RegisterShape from public API. Use C++ shape function registration instead.
indexing now starts from 1 instead of 0, and `bus_id==0` is used where
previously `BUS_ANY` was used.
* Most RNN cells and RNN functions now use different variable scopes to be
consistent with layers (`tf.contrib.layers`). This means old checkpoints
written using this code will not load after this change without providing
@ -27,6 +68,65 @@ BUS_ANY was used.
* Deprecated tf.select op. tf.where should be used instead.
* `SparseTensor.shape` has been renamed to `SparseTensor.dense_shape`. Same for
`SparseTensorValue.shape`.
* `Env::FileExists` and `FileSystem::FileExists` now return a
`tensorflow::Status` intead of a bool. Any callers to this function can be
converted to a bool by adding `.ok()` to the call.
* C API: Type `TF_SessionWithGraph` has been renamed to `TF_Session`, indicating
its preferred use in language bindings for TensorFlow. What was previously
`TF_Session` has been renamed to `TF_DeprecatedSession`.
* C API: Renamed `TF_Port` to `TF_Output`.
* C API: The caller retains ownership of `TF_Tensor` objects provided to
`TF_Run`, `TF_SessionRun`, `TF_SetAttrTensor` etc.
* Renamed `tf.image.per_image_whitening()` to
`tf.image.per_image_standardization()`
* Move Summary protobuf constructors to `tf.summary` submodule.
* Deprecate `histogram_summary`, `audio_summary`, `scalar_summary`,
`image_summary`, `merge_summary`, and `merge_all_summaries`.
* Combined `batch_*` and regular version of linear algebra and FFT ops. The
regular op now handles batches as well. All `batch_*` Python interfaces were
removed.
* `tf.all_variables`, `tf.VARIABLES` and `tf.initialize_all_variables` renamed
to `tf.global_variables`, `tf.GLOBAL_VARIABLES` and
`tf.global_variables_initializer` respectively.
## Bug Fixes and Other Changes
* Use threadsafe version of `lgamma` function.
* Fix `tf.sqrt` handling of negative arguments.
* Fixed bug causing incorrect number of threads to be used for multi-threaded
benchmarks.
* Performance optimizations for `batch_matmul` on multi-core CPUs.
* Improve trace, `matrix_set_diag`, `matrix_diag_part` and their gradients to
work for rectangular matrices.
* Support for SVD of complex valued matrices.
## Thanks to our Contributors
This release contains contributions from many people at Google, as well as:
@a7744hsc, Abhi Agg, @admcrae, Adriano Carmezim, Aki Sukegawa, Alex Kendall,
Alexander Rosenberg Johansen, @amcrae, Amlan Kar, Andre Simpelo, Andreas Eberle,
Andrew Hundt, Arnaud Lenglet, @b0noI, Balachander Ramachandran, Ben Barsdell,
Ben Guidarelli, Benjamin Mularczyk, Burness Duan, @c0g, Changming Sun,
@chanis, Corey Wharton, Dan J, Daniel Trebbien, Darren Garvey, David Brailovsky,
David Jones, Di Zeng, @DjangoPeng, Dr. Kashif Rasul, @drag0, Fabrizio (Misto)
Milo, FabríCio Ceschin, @fp, @Ghedeon, @guschmue, Gökçen Eraslan, Haosdent
Huang, Haroen Viaene, Harold Cooper, Henrik Holst, @hoangmit, Ivan Ukhov, Javier
Dehesa, Jingtian Peng, Jithin Odattu, Joan Pastor, Johan Mathe, Johannes Mayer,
Jongwook Choi, Justus Schwabedal, Kai Wolf, Kamil Hryniewicz, Kamran Amini,
Karen Brems, Karl Lattimer, @kborer, Ken Shirriff, Kevin Rose, Larissa Laich,
Laurent Mazare, Leonard Lee, Liang-Chi Hsieh, Liangliang He, Luke Iwanski,
Marek Kolodziej, Moustafa Alzantot, @MrQianjinsi, @nagachika, Neil Han, Nick
Meehan, Niels Ole Salscheider, Nikhil Mishra, @nschuc, Ondrej Skopek, OndřEj
Filip, @OscarDPan, Pablo Moyano, Przemyslaw Tredak, @qitaishui, @Quarazy,
@raix852, Philipp Helo, Sam Abrahams, @SriramRamesh, Till Hoffmann, Tushar Soni,
@tvn, @tyfkda, Uwe Schmidt, Victor Villas, Vit Stepanovs, Vladislav Gubarev,
@wujingyue, Xuesong Yang, Yi Liu, Yilei Yang, @youyou3, Yuan (Terry) Tang,
Yuming Wang, Zafar Takhirov, @zhongyuk, Ziming Dong, @guotong1988
We are also grateful to all who filed issues or helped resolve them, asked and
answered questions, and were part of inspiring discussions.
# Release 0.11.0

63
configure vendored
View File

@ -52,7 +52,6 @@ done
if is_windows; then
TF_NEED_GCP=0
TF_NEED_HDFS=0
TF_NEED_CUDA=0
TF_NEED_OPENCL=0
fi
@ -80,10 +79,10 @@ if [ "$TF_NEED_GCP" == "1" ]; then
fi
# Update Bazel build configuration.
perl -pi -e "s,WITH_GCP_SUPPORT = (False|True),WITH_GCP_SUPPORT = True,s" tensorflow/core/platform/default/build_config.bzl
sed -i -e "s/WITH_GCP_SUPPORT = False/WITH_GCP_SUPPORT = True/" tensorflow/core/platform/default/build_config.bzl
else
# Update Bazel build configuration.
perl -pi -e "s,WITH_GCP_SUPPORT = (False|True),WITH_GCP_SUPPORT = False,s" tensorflow/core/platform/default/build_config.bzl
sed -i -e "s/WITH_GCP_SUPPORT = True/WITH_GCP_SUPPORT = False/" tensorflow/core/platform/default/build_config.bzl
fi
while [ "$TF_NEED_HDFS" == "" ]; do
@ -102,10 +101,10 @@ done
if [ "$TF_NEED_HDFS" == "1" ]; then
# Update Bazel build configuration.
perl -pi -e "s,WITH_HDFS_SUPPORT = (False|True),WITH_HDFS_SUPPORT = True,s" tensorflow/core/platform/default/build_config.bzl
sed -i -e "s/WITH_HDFS_SUPPORT = False/WITH_HDFS_SUPPORT = True/" tensorflow/core/platform/default/build_config.bzl
else
# Update Bazel build configuration.
perl -pi -e "s,WITH_HDFS_SUPPORT = (False|True),WITH_HDFS_SUPPORT = False,s" tensorflow/core/platform/default/build_config.bzl
sed -i -e "s/WITH_HDFS_SUPPORT = True/WITH_HDFS_SUPPORT = False/" tensorflow/core/platform/default/build_config.bzl
fi
# Invoke python_config and set up symlinks to python includes
@ -131,11 +130,11 @@ done
## Set up Cuda-related environment settings
while [ "$TF_NEED_CUDA" == "" ]; do
read -p "Do you wish to build TensorFlow with GPU support? [y/N] " INPUT
read -p "Do you wish to build TensorFlow with CUDA support? [y/N] " INPUT
case $INPUT in
[Yy]* ) echo "GPU support will be enabled for TensorFlow"; TF_NEED_CUDA=1;;
[Nn]* ) echo "No GPU support will be enabled for TensorFlow"; TF_NEED_CUDA=0;;
"" ) echo "No GPU support will be enabled for TensorFlow"; TF_NEED_CUDA=0;;
[Yy]* ) echo "CUDA support will be enabled for TensorFlow"; TF_NEED_CUDA=1;;
[Nn]* ) echo "No CUDA support will be enabled for TensorFlow"; TF_NEED_CUDA=0;;
"" ) echo "No CUDA support will be enabled for TensorFlow"; TF_NEED_CUDA=0;;
* ) echo "Invalid selection: " $INPUT;;
esac
done
@ -150,14 +149,15 @@ fi
if [ "$TF_NEED_CUDA" == "1" ]; then
# Set up which gcc nvcc should use as the host compiler
while true; do
# No need to set this on Windows
while ! is_windows && true; do
fromuser=""
if [ -z "$GCC_HOST_COMPILER_PATH" ]; then
default_gcc_host_compiler_path=$(which gcc || true)
read -p "Please specify which gcc should be used by nvcc as the host compiler. [Default is $default_gcc_host_compiler_path]: " GCC_HOST_COMPILER_PATH
fromuser="1"
if [ -z "$GCC_HOST_COMPILER_PATH" ]; then
GCC_HOST_COMPILER_PATH=$default_gcc_host_compiler_path
GCC_HOST_COMPILER_PATH="$default_gcc_host_compiler_path"
fi
fi
if [ -e "$GCC_HOST_COMPILER_PATH" ]; then
@ -178,16 +178,23 @@ OSNAME=`uname -s`
while true; do
# Configure the Cuda SDK version to use.
if [ -z "$TF_CUDA_VERSION" ]; then
read -p "Please specify the Cuda SDK version you want to use, e.g. 7.0. [Leave empty to use system default]: " TF_CUDA_VERSION
read -p "Please specify the CUDA SDK version you want to use, e.g. 7.0. [Leave empty to use system default]: " TF_CUDA_VERSION
fi
fromuser=""
if [ -z "$CUDA_TOOLKIT_PATH" ]; then
default_cuda_path=/usr/local/cuda
if is_windows; then
if [ -z "$CUDA_PATH" ]; then
default_cuda_path="C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v8.0"
else
default_cuda_path="$(cygpath -m "$CUDA_PATH")"
fi
fi
read -p "Please specify the location where CUDA $TF_CUDA_VERSION toolkit is installed. Refer to README.md for more details. [Default is $default_cuda_path]: " CUDA_TOOLKIT_PATH
fromuser="1"
if [ -z "$CUDA_TOOLKIT_PATH" ]; then
CUDA_TOOLKIT_PATH=$default_cuda_path
CUDA_TOOLKIT_PATH="$default_cuda_path"
fi
fi
@ -197,7 +204,9 @@ while true; do
TF_CUDA_EXT=".$TF_CUDA_VERSION"
fi
if [ "$OSNAME" == "Linux" ]; then
if is_windows; then
CUDA_RT_LIB_PATH="lib/x64/cudart.lib"
elif [ "$OSNAME" == "Linux" ]; then
CUDA_RT_LIB_PATH="lib64/libcudart.so${TF_CUDA_EXT}"
elif [ "$OSNAME" == "Darwin" ]; then
CUDA_RT_LIB_PATH="lib/libcudart${TF_CUDA_EXT}.dylib"
@ -235,14 +244,17 @@ while true; do
fi
# Result returned from "read" will be used unexpanded. That make "~" unuseable.
# Going through one more level of expansion to handle that.
CUDNN_INSTALL_PATH=`${PYTHON_BIN_PATH} -c "import os; print(os.path.realpath(os.path.expanduser('${CUDNN_INSTALL_PATH}')))"`
CUDNN_INSTALL_PATH=`"${PYTHON_BIN_PATH}" -c "import os; print(os.path.realpath(os.path.expanduser('${CUDNN_INSTALL_PATH}')))"`
fi
if [[ -z "$TF_CUDNN_VERSION" ]]; then
TF_CUDNN_EXT=""
cudnn_lib_path=""
cudnn_alt_lib_path=""
if [ "$OSNAME" == "Linux" ]; then
if is_windows; then
cudnn_lib_path="${CUDNN_INSTALL_PATH}/lib/x64/cudnn.lib"
cudnn_alt_lib_path="${CUDNN_INSTALL_PATH}/lib/x64/cudnn.lib"
elif [ "$OSNAME" == "Linux" ]; then
cudnn_lib_path="${CUDNN_INSTALL_PATH}/lib64/libcudnn.so"
cudnn_alt_lib_path="${CUDNN_INSTALL_PATH}/libcudnn.so"
elif [ "$OSNAME" == "Darwin" ]; then
@ -255,9 +267,9 @@ while true; do
# If the path is not a symlink, readlink will exit with an error code, so
# in that case, we return the path itself.
if [ -f "$cudnn_lib_path" ]; then
REALVAL=`readlink ${cudnn_lib_path} || echo "${cudnn_lib_path}"`
REALVAL=`readlink "${cudnn_lib_path}" || echo "${cudnn_lib_path}"`
else
REALVAL=`readlink ${cudnn_alt_lib_path} || echo "${cudnn_alt_lib_path}"`
REALVAL=`readlink "${cudnn_alt_lib_path}" || echo "${cudnn_alt_lib_path}"`
fi
# Extract the version of the SONAME, if it was indeed symlinked to
@ -275,7 +287,10 @@ while true; do
TF_CUDNN_EXT=".$TF_CUDNN_VERSION"
fi
if [ "$OSNAME" == "Linux" ]; then
if is_windows; then
CUDA_DNN_LIB_PATH="lib/x64/cudnn.lib"
CUDA_DNN_LIB_ALT_PATH="lib/x64/cudnn.lib"
elif [ "$OSNAME" == "Linux" ]; then
CUDA_DNN_LIB_PATH="lib64/libcudnn.so${TF_CUDNN_EXT}"
CUDA_DNN_LIB_ALT_PATH="libcudnn.so${TF_CUDNN_EXT}"
elif [ "$OSNAME" == "Darwin" ]; then
@ -350,6 +365,16 @@ EOF
TF_CUDA_COMPUTE_CAPABILITIES=""
done
if is_windows; then
# The following three variables are needed for MSVC toolchain configuration in Bazel
export CUDA_PATH="$CUDA_TOOLKIT_PATH"
export CUDA_COMPUTE_CAPABILITIES="$TF_CUDA_COMPUTE_CAPABILITIES"
export NO_WHOLE_ARCHIVE_OPTION=1
# Set GCC_HOST_COMPILER_PATH to keep cuda_configure.bzl happy
export GCC_HOST_COMPILER_PATH="/usr/bin/dummy_compiler"
fi
# end of if "$TF_NEED_CUDA" == "1"
fi

3
grpc.BUILD
View File

@ -1561,6 +1561,7 @@ cc_library(
"include/grpc++/impl/codegen/create_auth_context.h",
"include/grpc++/impl/codegen/grpc_library.h",
"include/grpc++/impl/codegen/method_handler_impl.h",
"include/grpc++/impl/codegen/proto_utils.h",
"include/grpc++/impl/codegen/rpc_method.h",
"include/grpc++/impl/codegen/rpc_service_method.h",
"include/grpc++/impl/codegen/security/auth_context.h",
@ -1766,10 +1767,12 @@ cc_library(
"include/grpc++/impl/codegen/completion_queue.h",
"include/grpc++/impl/codegen/completion_queue_tag.h",
"include/grpc++/impl/codegen/config.h",
"include/grpc++/impl/codegen/config_protobuf.h",
"include/grpc++/impl/codegen/core_codegen_interface.h",
"include/grpc++/impl/codegen/create_auth_context.h",
"include/grpc++/impl/codegen/grpc_library.h",
"include/grpc++/impl/codegen/method_handler_impl.h",
"include/grpc++/impl/codegen/proto_utils.h",
"include/grpc++/impl/codegen/rpc_method.h",
"include/grpc++/impl/codegen/rpc_service_method.h",
"include/grpc++/impl/codegen/security/auth_context.h",

9
tensorflow/cc/gradients/nn_grad_test.cc
View File

@ -63,7 +63,8 @@ TEST_F(NNGradTest, ReluGrad) {
auto y = Relu(scope_, x);
// Avoid input values where ReLU gradient is not well defined (around zero).
Tensor x_init_value = test::AsTensor<float>(
{-0.9, -0.7, -0.5, -0.3, -0.1, 0.1, 0.3, 0.5, 0.7, 0.9}, {5, 2});
{-0.9f, -0.7f, -0.5f, -0.3f, -0.1f, 0.1f, 0.3f, 0.5f, 0.7f, 0.9f},
{5, 2});
RunTest(x, x_init_value, y, shape);
}
@ -74,7 +75,8 @@ TEST_F(NNGradTest, Relu6Grad) {
// Avoid input values where ReLU gradient is not well defined (around zero
// and six).
Tensor x_init_value = test::AsTensor<float>(
{-0.9, -0.7, -0.5, -0.3, -0.1, 6.1, 6.3, 6.5, 6.7, 6.9}, {5, 2});
{-0.9f, -0.7f, -0.5f, -0.3f, -0.1f, 6.1f, 6.3f, 6.5f, 6.7f, 6.9f},
{5, 2});
RunTest(x, x_init_value, y, shape);
}
@ -83,7 +85,8 @@ TEST_F(NNGradTest, EluGrad) {
auto x = Placeholder(scope_, DT_FLOAT, Placeholder::Shape(shape));
auto y = Elu(scope_, x);
Tensor x_init_value = test::AsTensor<float>(
{-0.9, -0.7, -0.5, -0.3, -0.1, 0.1, 0.3, 0.5, 0.7, 0.9}, {5, 2});
{-0.9f, -0.7f, -0.5f, -0.3f, -0.1f, 0.1f, 0.3f, 0.5f, 0.7f, 0.9f},
{5, 2});
RunTest(x, x_init_value, y, shape);
}

5
tensorflow/contrib/bayesflow/python/kernel_tests/stochastic_tensor_test.py
View File

@ -229,3 +229,8 @@ class ObservedStochasticTensorTest(tf.test.TestCase):
distributions.Normal(mu=mu, sigma=sigma),
value=tf.zeros(
(1, 2), dtype=tf.int32))
if __name__ == "__main__":
tf.test.main()

8
tensorflow/contrib/bayesflow/python/kernel_tests/stochastic_variables_test.py
View File

@ -42,8 +42,8 @@ class StochasticVariablesTest(tf.test.TestCase):
self.assertEqual(
{"stochastic_variables/sv_mu", "stochastic_variables/sv_sigma"},
set([v.op.name for v in tf.all_variables()]))
self.assertEqual(set(tf.trainable_variables()), set(tf.all_variables()))
set([v.op.name for v in tf.global_variables()]))
self.assertEqual(set(tf.trainable_variables()), set(tf.global_variables()))
v = tf.convert_to_tensor(v)
self.assertEqual(list(shape), v.get_shape().as_list())
@ -64,7 +64,7 @@ class StochasticVariablesTest(tf.test.TestCase):
})):
v = tf.get_variable("sv")
for var in tf.all_variables():
for var in tf.global_variables():
if "mu" in var.name:
mu_var = var
if "sigma" in var.name:
@ -96,7 +96,7 @@ class StochasticVariablesTest(tf.test.TestCase):
})):
v = tf.get_variable("sv", shape)
for var in tf.all_variables():
for var in tf.global_variables():
if "mu" in var.name:
mu_var = var
if "sigma" in var.name:

18
tensorflow/contrib/cmake/CMakeLists.txt
View File

@ -1,5 +1,5 @@
# Minimum CMake required
cmake_minimum_required(VERSION 3.1)
cmake_minimum_required(VERSION 3.5)
# Project
project(tensorflow C CXX)
@ -28,9 +28,11 @@ option(tensorflow_BUILD_CONTRIB_KERNELS "Build OpKernels from tensorflow/contrib
option(tensorflow_BUILD_CC_TESTS "Build cc unit tests " OFF)
option(tensorflow_BUILD_PYTHON_TESTS "Build python unit tests " OFF)
#Threads: defines CMAKE_THREAD_LIBS_INIT and adds -pthread compile option for
# targets that link ${CMAKE_THREAD_LIBS_INIT}.
find_package (Threads)
if (NOT WIN32)
# Threads: defines CMAKE_THREAD_LIBS_INIT and adds -pthread compile option
# for targets that link ${CMAKE_THREAD_LIBS_INIT}.
find_package (Threads)
endif()
# [CLEANUP] Remove when done
# For debugging
@ -55,6 +57,7 @@ if(WIN32)
add_definitions(-DNOMINMAX -D_WIN32_WINNT=0x0A00 -DLANG_CXX11 -DCOMPILER_MSVC -D__VERSION__=\"MSVC\")
add_definitions(-DWIN32 -DOS_WIN -D_MBCS -DWIN64 -DWIN32_LEAN_AND_MEAN -DNOGDI -DPLATFORM_WINDOWS)
add_definitions(-DTENSORFLOW_USE_EIGEN_THREADPOOL -DEIGEN_HAS_C99_MATH -D_ITERATOR_DEBUG_LEVEL=0)
add_definitions(-DNDEBUG /O2) # Equivalent of -c opt in Bazel.
add_definitions(/bigobj /nologo /EHsc /GF /FC /MP /Gm-)
# Suppress warnings to reduce build log size.
add_definitions(/wd4267 /wd4244 /wd4800 /wd4503 /wd4554 /wd4996 /wd4348 /wd4018)
@ -147,11 +150,11 @@ if (tensorflow_ENABLE_GPU)
# by default we assume compute cabability 3.5 and 5.2. If you change this change it in
# CUDA_NVCC_FLAGS and cuda_config.h below
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};-gencode arch=compute_35,code=\"sm_35,compute_35\";-gencode arch=compute_52,code=\"sm_52,compute_52\")
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};-gencode arch=compute_30,code=\"sm_30,compute_30\";-gencode arch=compute_35,code=\"sm_35,compute_35\";-gencode arch=compute_52,code=\"sm_52,compute_52\")
set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--include-path ${PROJECT_BINARY_DIR}/$\{build_configuration\};--expt-relaxed-constexpr)
set(CUDA_INCLUDE ${CUDA_TOOLKIT_TARGET_DIR} ${CUDA_TOOLKIT_TARGET_DIR}/extras/CUPTI/include)
include_directories(${CUDA_INCLUDE})
add_definitions(-DGOOGLE_CUDA=1 -DTF_EXTRA_CUDA_CAPABILITIES=3.5,5.2)
add_definitions(-DGOOGLE_CUDA=1 -DTF_EXTRA_CUDA_CAPABILITIES=3.0,3.5,5.2)
# add cudnn
include_directories(${CUDNN_HOME})
@ -161,7 +164,7 @@ if (tensorflow_ENABLE_GPU)
FILE(WRITE ${tensorflow_source_dir}/third_party/gpus/cuda/cuda_config.h
"#ifndef CUDA_CUDA_CONFIG_H_\n"
"#define CUDA_CUDA_CONFIG_H_\n"
"#define TF_CUDA_CAPABILITIES CudaVersion(\"3.5\"),CudaVersion(\"5.2\")\n"
"#define TF_CUDA_CAPABILITIES CudaVersion(\"3.0\"),CudaVersion(\"3.5\"),CudaVersion(\"5.2\")\n"
"#define TF_CUDA_VERSION \"64_80\"\n"
"#define TF_CUDNN_VERSION \"64_5\"\n"
"#endif // CUDA_CUDA_CONFIG_H_\n"
@ -207,6 +210,7 @@ if(tensorflow_BUILD_CC_EXAMPLE)
include(tf_label_image_example.cmake)
endif()
if(tensorflow_BUILD_PYTHON_BINDINGS)
include(tensorboard)
include(tf_python.cmake)
endif()
if (tensorflow_BUILD_CC_TESTS OR tensorflow_BUILD_PYTHON_TESTS)

2
tensorflow/contrib/cmake/README.md
View File

@ -21,7 +21,7 @@ Note: Windows support is in an **alpha** state, and we welcome your feedback.
### Pre-requisites
* CMake version 3.1 up to 3.6
* CMake version 3.5 up to 3.6
* [Git](http://git-scm.com)

134
tensorflow/contrib/cmake/external/tensorboard.cmake vendored Normal file
View File

@ -0,0 +1,134 @@
include (ExternalProject)
set(tensorboard_dependencies)
add_custom_target(tensorboard_copy_dependencies)
function(tb_new_http_archive)
cmake_parse_arguments(_TB "" "NAME;URL" "FILES" ${ARGN})
ExternalProject_Add(${_TB_NAME}
PREFIX ${_TB_NAME}
URL ${_TB_URL}
DOWNLOAD_DIR "${DOWNLOAD_LOCATION}/${_TB_NAME}"
CONFIGURE_COMMAND ""
BUILD_COMMAND ""
INSTALL_COMMAND ""
)
set(src_dir "${CMAKE_CURRENT_BINARY_DIR}/${_TB_NAME}/src/${_TB_NAME}")
set(dst_dir "${CMAKE_CURRENT_BINARY_DIR}/tensorboard_external/${_TB_NAME}")
foreach(src_file ${_TB_FILES})
add_custom_command(
TARGET tensorboard_copy_dependencies PRE_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${src_dir}/${src_file} ${dst_dir}/${src_file}
)
endforeach()
set(tensorboard_dependencies ${tensorboard_dependencies} ${_TB_NAME} PARENT_SCOPE)
endfunction()
function(tb_http_file)
cmake_parse_arguments(_TB "" "NAME;URL" "" ${ARGN})
get_filename_component(src_file ${_TB_URL} NAME)
file(DOWNLOAD ${_TB_URL} "${DOWNLOAD_LOCATION}/${_TB_NAME}/${src_file}")
set(src_dir "${DOWNLOAD_LOCATION}/${_TB_NAME}")
set(dst_dir "${CMAKE_CURRENT_BINARY_DIR}/tensorboard_external/${_TB_NAME}/file")
add_custom_command(
TARGET tensorboard_copy_dependencies PRE_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${src_dir}/${src_file} ${dst_dir}/${src_file}
)
add_custom_target(${_TB_NAME} DEPENDS ${src_dir}/${src_file})
set(tensorboard_dependencies ${tensorboard_dependencies} ${_TB_NAME} PARENT_SCOPE)
endfunction()
# Parse TensorBoard dependency names and URLs from Bazel's WORKSPACE file.
set(tb_dep_names)
file(STRINGS ${PROJECT_SOURCE_DIR}/../../../WORKSPACE workspace_contents)
foreach(line ${workspace_contents})
if(line MATCHES "# TENSORBOARD_BOWER_AUTOGENERATED_BELOW_THIS_LINE_DO_NOT_EDIT")
set(tb_deps_started 1)
endif()
if(NOT tb_deps_started)
continue()
endif()
if(line MATCHES "new_http_archive\\(")
set(tb_dep_is_archive 1)
continue()
elseif(line MATCHES "http_file\\(")
set(tb_dep_is_archive 0)
continue()
endif()
string(REGEX MATCH "name.*=.*\"(.*)\"" has_name ${line})
if(has_name)
set(tb_dep_name ${CMAKE_MATCH_1})
continue()
endif()
string(REGEX MATCH "url.*=.*\"(.*)\"" has_url ${line})
if(has_url)
list(APPEND tb_dep_names ${tb_dep_name})
set(${tb_dep_name}_is_archive ${tb_dep_is_archive})
set(${tb_dep_name}_url ${CMAKE_MATCH_1})
endif()
endforeach()
# Parse the files needed for each TensorBoard dependency from Bazel's bower.BUILD file.
# Due to CMAKE quirkiness, cannot use file(strings) with files that contain '[' and ']'.
file(READ ${PROJECT_SOURCE_DIR}/../../../bower.BUILD bower_build_contents)
string(REPLACE "\[" "OB" bower_build_contents "${bower_build_contents}")
string(REPLACE "\]" "CB" bower_build_contents "${bower_build_contents}")
string(REPLACE ";" "\\\\;" bower_build_contents "${bower_build_contents}")
string(REPLACE "\n" "E;" bower_build_contents "${bower_build_contents}")
foreach(line ${bower_build_contents})
string(REGEX MATCH "name.*=.*\"(.*)\"" has_name ${line})
if(has_name)
set(tb_dep_name ${CMAKE_MATCH_1})
set(${tb_dep_name}_files)
continue()
endif()
string(REGEX MATCH "srcs.*=.*\"(.*)\"CB" has_single_line_src ${line})
if(has_single_line_src)
list(APPEND ${tb_dep_name}_files ${CMAKE_MATCH_1})
continue()
endif()
if(line MATCHES "srcs.*=.*OB")
set(inside_files_def 1)
continue()
elseif(line MATCHES "CB,")
set(inside_files_def 0)
continue()
endif()
if(inside_files_def)
string(REGEX MATCH "\"(.*)\"," has_file ${line})
if(has_file)
list(APPEND ${tb_dep_name}_files ${CMAKE_MATCH_1})
endif()
endif()
endforeach()
# Generate a target for each dependency.
foreach(tb_dep_name ${tb_dep_names})
if (${tb_dep_name}_is_archive)
tb_new_http_archive(
NAME ${tb_dep_name}
URL ${${tb_dep_name}_url}
FILES ${${tb_dep_name}_files}
)
else()
tb_http_file(
NAME ${tb_dep_name}
URL ${${tb_dep_name}_url}
)
endif()
endforeach()
add_dependencies(tensorboard_copy_dependencies ${tensorboard_dependencies})

2
tensorflow/contrib/cmake/setup.py
View File

@ -26,7 +26,7 @@ from setuptools import find_packages, setup, Command
from setuptools.command.install import install as InstallCommandBase
from setuptools.dist import Distribution
_VERSION = '0.11.0-cmake-experimental'
_VERSION = '0.12.0-rc0-cmake-experimental'
REQUIRED_PACKAGES = [
'numpy >= 1.11.0',

33
tensorflow/contrib/cmake/tf_cc_ops.cmake
View File

@ -84,3 +84,36 @@ add_library(tf_cc_ops OBJECT
"${tensorflow_source_dir}/tensorflow/cc/ops/const_op.cc"
"${tensorflow_source_dir}/tensorflow/cc/ops/standard_ops.h"
)
########################################################
# tf_cc library
########################################################
file(GLOB_RECURSE tf_cc_srcs
"${tensorflow_source_dir}/tensorflow/cc/client/*.h"
"${tensorflow_source_dir}/tensorflow/cc/client/*.cc"
"${tensorflow_source_dir}/tensorflow/cc/gradients/*.h"
"${tensorflow_source_dir}/tensorflow/cc/gradients/*.cc"
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*.h"
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*.cc"
"${tensorflow_source_dir}/tensorflow/cc/training/*.h"
"${tensorflow_source_dir}/tensorflow/cc/training/*.cc"
)
set(tf_cc_srcs
${tf_cc_srcs}
"${tensorflow_source_dir}/tensorflow/cc/framework/grad_op_registry.h"
"${tensorflow_source_dir}/tensorflow/cc/framework/grad_op_registry.cc"
"${tensorflow_source_dir}/tensorflow/cc/framework/gradient_checker.h"
"${tensorflow_source_dir}/tensorflow/cc/framework/gradient_checker.cc"
"${tensorflow_source_dir}/tensorflow/cc/framework/gradients.h"
"${tensorflow_source_dir}/tensorflow/cc/framework/gradients.cc"
)
file(GLOB_RECURSE tf_cc_test_srcs
"${tensorflow_source_dir}/tensorflow/cc/*test*.cc"
)
list(REMOVE_ITEM tf_cc_srcs ${tf_cc_test_srcs})
add_library(tf_cc OBJECT ${tf_cc_srcs})
add_dependencies(tf_cc tf_cc_framework tf_cc_ops)

6
tensorflow/contrib/cmake/tf_core_framework.cmake
View File

@ -210,11 +210,7 @@ file(GLOB_RECURSE tf_core_framework_test_srcs
"${tensorflow_source_dir}/tensorflow/core/util/*main.cc"
)
list(REMOVE_ITEM tf_core_framework_srcs ${tf_core_framework_test_srcs}
"${tensorflow_source_dir}/tensorflow/core/util/memmapped_file_system.cc"
"${tensorflow_source_dir}/tensorflow/core/util/memmapped_file_system.h"
"${tensorflow_source_dir}/tensorflow/core/util/memmapped_file_system_writer.cc"
)
list(REMOVE_ITEM tf_core_framework_srcs ${tf_core_framework_test_srcs})
add_library(tf_core_framework OBJECT
${tf_core_framework_srcs}

2
tensorflow/contrib/cmake/tf_core_kernels.cmake
View File

@ -90,8 +90,6 @@ list(REMOVE_ITEM tf_core_kernels_srcs ${tf_core_kernels_exclude_srcs})
if(WIN32)
file(GLOB_RECURSE tf_core_kernels_windows_exclude_srcs
# not working on windows yet
"${tensorflow_source_dir}/tensorflow/core/kernels/depthwise_conv_op.cc" # Cannot find symbol: tensorflow::LaunchConv2DOp<struct Eigen::ThreadPoolDevice, double>::launch(...).
"${tensorflow_source_dir}/tensorflow/core/kernels/fact_op.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/meta_support.*"
"${tensorflow_source_dir}/tensorflow/core/kernels/*quantiz*.h"
"${tensorflow_source_dir}/tensorflow/core/kernels/*quantiz*.cc"

2
tensorflow/contrib/cmake/tf_core_ops.cmake
View File

@ -9,6 +9,7 @@ set(tf_op_lib_names
"io_ops"
"linalg_ops"
"logging_ops"
"losses"
"math_ops"
"nn_ops"
"no_op"
@ -17,6 +18,7 @@ set(tf_op_lib_names
"resource_variable_ops"
"script_ops"
"sdca_ops"
"set_ops"
"sendrecv_ops"
"sparse_ops"
"state_ops"

99
tensorflow/contrib/cmake/tf_python.cmake
View File

@ -68,7 +68,7 @@ function(RELATIVE_PROTOBUF_GENERATE_PYTHON ROOT_DIR SRCS)
add_custom_command(
OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/tf_python/${REL_DIR}/${FIL_WE}_pb2.py"
COMMAND ${PROTOBUF_PROTOC_EXECUTABLE}
ARGS --python_out ${CMAKE_CURRENT_BINARY_DIR}/tf_python/ -I ${ROOT_DIR} -I ${PROTOBUF_INCLUDE_DIRS} ${ABS_FIL}
ARGS --python_out ${CMAKE_CURRENT_BINARY_DIR}/tf_python/ -I ${ROOT_DIR} -I ${PROTOBUF_INCLUDE_DIRS} ${ABS_FIL}
DEPENDS ${PROTOBUF_PROTOC_EXECUTABLE} protobuf
COMMENT "Running Python protocol buffer compiler on ${FIL}"
VERBATIM )
@ -118,12 +118,20 @@ RELATIVE_PROTOBUF_GENERATE_PYTHON(
${tensorflow_source_dir} PYTHON_PROTO_GENFILES ${tf_protos_python_srcs}
)
# NOTE(mrry): Avoid regenerating the tensorflow/core protos because this
# can cause benign-but-failing-on-Windows-due-to-file-locking conflicts
# when two rules attempt to generate the same file.
file(GLOB_RECURSE tf_python_protos_cc_srcs RELATIVE ${tensorflow_source_dir}
"${tensorflow_source_dir}/tensorflow/python/*.proto"
"${tensorflow_source_dir}/tensorflow/contrib/session_bundle/*.proto"
"${tensorflow_source_dir}/tensorflow/contrib/tensorboard/*.proto"
)
RELATIVE_PROTOBUF_GENERATE_CPP(PROTO_SRCS PROTO_HDRS
${tensorflow_source_dir} ${tf_protos_python_srcs}
${tensorflow_source_dir} ${tf_python_protos_cc_srcs}
)
add_library(tf_python_protos_cc ${PROTO_SRCS} ${PROTO_HDRS})
add_dependencies(tf_python_protos_cc tf_protos_cc)
# tf_python_touchup_modules adds empty __init__.py files to all
# directories containing Python code, so that Python will recognize
@ -141,7 +149,7 @@ function(add_python_module MODULE_NAME)
set(options DONTCOPY)
cmake_parse_arguments(ADD_PYTHON_MODULE "${options}" "" "" ${ARGN})
add_custom_command(TARGET tf_python_touchup_modules PRE_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory "${CMAKE_CURRENT_BINARY_DIR}/tf_python/${MODULE_NAME}")
COMMAND ${CMAKE_COMMAND} -E make_directory "${CMAKE_CURRENT_BINARY_DIR}/tf_python/${MODULE_NAME}")
add_custom_command(TARGET tf_python_touchup_modules PRE_BUILD
COMMAND ${CMAKE_COMMAND} -E touch "${CMAKE_CURRENT_BINARY_DIR}/tf_python/${MODULE_NAME}/__init__.py")
file(GLOB module_python_srcs RELATIVE ${tensorflow_source_dir}
@ -164,6 +172,9 @@ add_python_module("tensorflow/core/lib")
add_python_module("tensorflow/core/lib/core")
add_python_module("tensorflow/core/protobuf")
add_python_module("tensorflow/core/util")
add_python_module("tensorflow/examples")
add_python_module("tensorflow/examples/tutorials")
add_python_module("tensorflow/examples/tutorials/mnist")
add_python_module("tensorflow/python")
add_python_module("tensorflow/python/client")
add_python_module("tensorflow/python/debug")
@ -172,6 +183,7 @@ add_python_module("tensorflow/python/debug/examples")
add_python_module("tensorflow/python/debug/wrappers")
add_python_module("tensorflow/python/framework")
add_python_module("tensorflow/python/kernel_tests")
add_python_module("tensorflow/python/layers")
add_python_module("tensorflow/python/lib")
add_python_module("tensorflow/python/lib/core")
add_python_module("tensorflow/python/lib/io")
@ -179,6 +191,7 @@ add_python_module("tensorflow/python/ops")
add_python_module("tensorflow/python/platform")
add_python_module("tensorflow/python/platform/default")
add_python_module("tensorflow/python/platform/summary")
add_python_module("tensorflow/python/saved_model")
add_python_module("tensorflow/python/summary")
add_python_module("tensorflow/python/summary/impl")
add_python_module("tensorflow/python/summary/writer")
@ -187,8 +200,13 @@ add_python_module("tensorflow/python/training")
add_python_module("tensorflow/python/user_ops")
add_python_module("tensorflow/python/util")
add_python_module("tensorflow/python/util/protobuf")
add_python_module("tensorflow/contrib/")
add_python_module("tensorflow/tensorboard")
add_python_module("tensorflow/tensorboard/backend")
add_python_module("tensorflow/tensorboard/lib/python")
add_python_module("tensorflow/tensorboard/plugins")
add_python_module("tensorflow/tensorboard/plugins/projector")
add_python_module("tensorflow/tensorboard/scripts")
add_python_module("tensorflow/contrib")
add_python_module("tensorflow/contrib/android")
add_python_module("tensorflow/contrib/android/java")
add_python_module("tensorflow/contrib/android/java/org")
@ -215,6 +233,7 @@ add_python_module("tensorflow/contrib/cudnn_rnn/ops")
add_python_module("tensorflow/contrib/cudnn_rnn/python")
add_python_module("tensorflow/contrib/cudnn_rnn/python/kernel_tests")
add_python_module("tensorflow/contrib/cudnn_rnn/python/ops")
add_python_module("tensorflow/contrib/deprecated")
add_python_module("tensorflow/contrib/distributions")
add_python_module("tensorflow/contrib/distributions/python")
add_python_module("tensorflow/contrib/distributions/python/kernel_tests")
@ -256,6 +275,9 @@ add_python_module("tensorflow/contrib/ios_examples/camera/en.lproj")
add_python_module("tensorflow/contrib/ios_examples/simple")
add_python_module("tensorflow/contrib/ios_examples/simple/data")
add_python_module("tensorflow/contrib/ios_examples/simple/tf_ios_makefile_example.xcodeproj")
add_python_module("tensorflow/contrib/labeled_tensor")
add_python_module("tensorflow/contrib/labeled_tensor/python")
add_python_module("tensorflow/contrib/labeled_tensor/python/ops")
add_python_module("tensorflow/contrib/layers")
add_python_module("tensorflow/contrib/layers/kernels")
add_python_module("tensorflow/contrib/layers/ops")
@ -279,6 +301,10 @@ add_python_module("tensorflow/contrib/learn/python/learn/preprocessing/tests")
add_python_module("tensorflow/contrib/learn/python/learn/tests")
add_python_module("tensorflow/contrib/learn/python/learn/tests/dataframe")
add_python_module("tensorflow/contrib/learn/python/learn/utils")
add_python_module("tensorflow/contrib/linalg")
add_python_module("tensorflow/contrib/linalg/python")
add_python_module("tensorflow/contrib/linalg/python/ops")
add_python_module("tensorflow/contrib/linalg/python/kernel_tests")
add_python_module("tensorflow/contrib/linear_optimizer")
add_python_module("tensorflow/contrib/linear_optimizer/kernels")
add_python_module("tensorflow/contrib/linear_optimizer/kernels/g3doc")
@ -329,8 +355,12 @@ add_python_module("tensorflow/contrib/slim/python")
add_python_module("tensorflow/contrib/slim/python/slim")
add_python_module("tensorflow/contrib/slim/python/slim/data")
add_python_module("tensorflow/contrib/slim/python/slim/nets")
add_python_module("tensorflow/contrib/solvers")
add_python_module("tensorflow/contrib/solvers/python")
add_python_module("tensorflow/contrib/solvers/python/ops")
add_python_module("tensorflow/contrib/specs")
add_python_module("tensorflow/contrib/specs/python")
add_python_module("tensorflow/contrib/stat_summarizer")
add_python_module("tensorflow/contrib/tensorboard")
add_python_module("tensorflow/contrib/tensorboard/plugins")
add_python_module("tensorflow/contrib/tensorboard/plugins/projector")
@ -350,10 +380,6 @@ add_python_module("tensorflow/contrib/tensor_forest/hybrid/python/ops")
add_python_module("tensorflow/contrib/tensor_forest/python")
add_python_module("tensorflow/contrib/tensor_forest/python/kernel_tests")
add_python_module("tensorflow/contrib/tensor_forest/python/ops")
add_python_module("tensorflow/contrib/tensorboard")
add_python_module("tensorflow/contrib/tensorboard")
add_python_module("tensorflow/contrib/tensorboard/plugins")
add_python_module("tensorflow/contrib/tensorboard/plugins/projector")
add_python_module("tensorflow/contrib/testing")
add_python_module("tensorflow/contrib/testing/python")
add_python_module("tensorflow/contrib/testing/python/framework")
@ -366,6 +392,12 @@ add_python_module("tensorflow/contrib/training/python")
add_python_module("tensorflow/contrib/training/python/training")
add_python_module("tensorflow/contrib/util")
# Additional directories with no Python sources.
add_custom_command(TARGET tf_python_touchup_modules PRE_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory "${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/tensorboard/dist")
add_custom_command(TARGET tf_python_touchup_modules PRE_BUILD
COMMAND ${CMAKE_COMMAND} -E make_directory "${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/tensorboard/lib/css")
########################################################
# tf_python_op_gen_main library
@ -394,7 +426,7 @@ function(GENERATE_PYTHON_OP_LIB tf_python_op_lib_name)
set(oneValueArgs DESTINATION)
set(multiValueArgs ADDITIONAL_LIBRARIES)
cmake_parse_arguments(GENERATE_PYTHON_OP_LIB
"" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
"${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
if(NOT DEFINED GENERATE_PYTHON_OP_LIB_DESTINATION)
# Default destination is tf_python/tensorflow/python/ops/gen_<...>.py.
set(GENERATE_PYTHON_OP_LIB_DESTINATION
@ -424,7 +456,7 @@ function(GENERATE_PYTHON_OP_LIB tf_python_op_lib_name)
COMMAND ${tf_python_op_lib_name}_gen_python @${tensorflow_source_dir}/tensorflow/python/ops/hidden_ops.txt 1 > ${GENERATE_PYTHON_OP_LIB_DESTINATION}
DEPENDS ${tf_python_op_lib_name}_gen_python
)
set(tf_python_ops_generated_files ${tf_python_ops_generated_files}
${GENERATE_PYTHON_OP_LIB_DESTINATION} PARENT_SCOPE)
endfunction()
@ -441,12 +473,14 @@ GENERATE_PYTHON_OP_LIB("image_ops")
GENERATE_PYTHON_OP_LIB("io_ops")
GENERATE_PYTHON_OP_LIB("linalg_ops")
GENERATE_PYTHON_OP_LIB("logging_ops")
GENERATE_PYTHON_OP_LIB("losses")
GENERATE_PYTHON_OP_LIB("nn_ops")
GENERATE_PYTHON_OP_LIB("parsing_ops")
GENERATE_PYTHON_OP_LIB("random_ops")
GENERATE_PYTHON_OP_LIB("resource_variable_ops")
GENERATE_PYTHON_OP_LIB("script_ops")
GENERATE_PYTHON_OP_LIB("sdca_ops")
GENERATE_PYTHON_OP_LIB("set_ops")
GENERATE_PYTHON_OP_LIB("state_ops")
GENERATE_PYTHON_OP_LIB("sparse_ops")
GENERATE_PYTHON_OP_LIB("string_ops")
@ -473,10 +507,13 @@ add_dependencies(tf_python_ops tf_python_op_gen_main)
find_package(SWIG REQUIRED)
# Generate the C++ and Python source code for the SWIG wrapper.
# NOTE(mrry): We always regenerate the SWIG wrapper, which means that we must
# always re-link the Python extension, but we don't have to track the
# individual headers on which the SWIG wrapper depends.
add_custom_command(
OUTPUT "${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/python/pywrap_tensorflow.py"
"${CMAKE_CURRENT_BINARY_DIR}/pywrap_tensorflow.cc"
DEPENDS tf_python_touchup_modules
DEPENDS tf_python_touchup_modules __force_rebuild
COMMAND ${SWIG_EXECUTABLE}
ARGS -python -c++
-I${tensorflow_source_dir}
@ -535,7 +572,7 @@ target_link_libraries(pywrap_tensorflow
${tf_core_gpu_kernels_lib}
${tensorflow_EXTERNAL_LIBRARIES}
tf_protos_cc
tf_python_protos_cc
tf_python_protos_cc
${PYTHON_LIBRARIES}
)
@ -545,11 +582,12 @@ target_link_libraries(pywrap_tensorflow
add_custom_target(tf_python_build_pip_package)
add_dependencies(tf_python_build_pip_package
pywrap_tensorflow
tensorboard_copy_dependencies
tf_python_copy_scripts_to_destination
tf_python_touchup_modules
tf_python_ops)
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/contrib/cmake/setup.py
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/tools/pip_package/setup.py
${CMAKE_CURRENT_BINARY_DIR}/tf_python/)
if(WIN32)
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
@ -566,6 +604,33 @@ add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/tools/pip_package/MANIFEST.in
${CMAKE_CURRENT_BINARY_DIR}/tf_python/)
# Copy resources for TensorBoard.
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/tf_python/setup.py bdist_wheel
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/tf_python)
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/tensorboard/dist/bazel-html-imports.html
${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/tensorboard/dist/)
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/tensorboard/dist/index.html
${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/tensorboard/dist/)
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/tensorboard/dist/tf-tensorboard.html
${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/tensorboard/dist/)
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/tensorboard/lib/css/global.css
${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/tensorboard/lib/css/)
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy ${tensorflow_source_dir}/tensorflow/tensorboard/TAG
${CMAKE_CURRENT_BINARY_DIR}/tf_python/tensorflow/tensorboard/)
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy_directory ${CMAKE_CURRENT_BINARY_DIR}/tensorboard_external
${CMAKE_CURRENT_BINARY_DIR}/tf_python/external)
if(${tensorflow_ENABLE_GPU})
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/tf_python/setup.py bdist_wheel --project_name tensorflow_gpu
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/tf_python)
else()
add_custom_command(TARGET tf_python_build_pip_package POST_BUILD
COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/tf_python/setup.py bdist_wheel
WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/tf_python)
endif(${tensorflow_ENABLE_GPU})

150
tensorflow/contrib/cmake/tf_tests.cmake
View File

@ -70,17 +70,18 @@ function(AddTest)
)
foreach(datafile ${_AT_DATA})
file(RELATIVE_PATH datafile_rel ${tensorflow_source_dir} ${datafile})
add_custom_command(
TARGET ${_AT_TARGET} POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy
"${CMAKE_CURRENT_SOURCE_DIR}/${datafile}"
"${testdir}/${datafile}"
DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/${datafile}"
"${datafile}"
"${testdir}/${datafile_rel}"
DEPENDS "${datafile}"
)
endforeach()
if (_AT_DEPENDS)
add_dependencies(${_AT_TARGET} ${_AT_DEPENDS})
add_dependencies(${_AT_TARGET} ${_AT_DEPENDS} googletest)
endif()
endfunction(AddTest)
@ -98,7 +99,7 @@ function(AddPythonTests)
endif(_AT_DEPENDS)
foreach(sourcefile ${_AT_SOURCES})
add_test(NAME ${sourcefile} COMMAND ${PYTHON_EXECUTABLE} ${sourcefile})
add_test(NAME ${sourcefile} COMMAND ${PYTHON_EXECUTABLE} ${sourcefile} WORKING_DIRECTORY ${tensorflow_source_dir})
if (_AT_DEPENDS)
add_dependencies(${_AT_TARGET} ${_AT_DEPENDS})
endif()
@ -116,6 +117,9 @@ if (tensorflow_BUILD_PYTHON_TESTS)
# include all test
file(GLOB_RECURSE tf_test_src_py
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/*.py"
"${tensorflow_source_dir}/tensorflow/python/saved_model/*_test.py"
"${tensorflow_source_dir}/tensorflow/python/training/*_test.py"
"${tensorflow_source_dir}/tensorflow/tensorboard/*_test.py"
)
# exclude the onces we don't want
@ -143,22 +147,21 @@ if (tensorflow_BUILD_PYTHON_TESTS)
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/cwise_ops_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/variable_scope_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/reshape_op_test.py"
"${tensorflow_source_dir}/tensorflow/tensorboard/backend/server_test.py"
# int32/int64 mixup
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/functional_ops_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/py_func_test.py"
# issues related to windows fs
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/io_ops_test.py"
# missing kernel
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/conv_ops_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/depthwise_conv_op_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/fractional_avg_pool_op_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/pool_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/qr_op_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/svd_op_test.py"
# cuda launch failed
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/diag_op_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/trace_op_test.py"
"${tensorflow_source_dir}/tensorflow/python/kernel_tests/one_hot_op_test.py" # gpu, T=uint8
# training tests
"${tensorflow_source_dir}/tensorflow/python/training/basic_session_run_hooks_test.py" # Needs tf.contrib fix.
"${tensorflow_source_dir}/tensorflow/python/training/localhost_cluster_performance_test.py" # Needs portpicker.
"${tensorflow_source_dir}/tensorflow/python/training/monitored_session_test.py" # Needs tf.contrib fix.
"${tensorflow_source_dir}/tensorflow/python/training/saver_large_variable_test.py" # Overflow error.
"${tensorflow_source_dir}/tensorflow/python/training/supervisor_test.py" # Flaky I/O error on rename.
"${tensorflow_source_dir}/tensorflow/python/training/sync_replicas_optimizer_test.py" # Needs portpicker.
)
endif()
list(REMOVE_ITEM tf_test_src_py ${tf_test_src_py_exclude})
@ -172,7 +175,7 @@ if (tensorflow_BUILD_CC_TESTS)
#
# cc unit tests. Be aware that by default we include 250+ tests which
# will take time and space to build.
# If you wan to cut this down, for example to a specific test, modify
# If you want to cut this down, for example to a specific test, modify
# tf_test_src_simple to your needs
#
@ -195,6 +198,18 @@ if (tensorflow_BUILD_CC_TESTS)
"${tensorflow_source_dir}/tensorflow/c/tf_status_helper.cc"
)
if(WIN32)
set(tf_src_testlib
${tf_src_testlib}
"${tensorflow_source_dir}/tensorflow/core/platform/windows/test.cc"
)
else()
set(tf_src_testlib
${tf_src_testlib}
"${tensorflow_source_dir}/tensorflow/core/platform/posix/test.cc"
)
endif()
# include all test
file(GLOB_RECURSE tf_test_src_simple
"${tensorflow_source_dir}/tensorflow/cc/*_test.cc"
@ -204,6 +219,15 @@ if (tensorflow_BUILD_CC_TESTS)
"${tensorflow_source_dir}/tensorflow/contrib/rnn/*_test.cc"
)
# exclude the ones we don't want
set(tf_test_src_simple_exclude
# generally not working
"${tensorflow_source_dir}/tensorflow/cc/client/client_session_test.cc"
"${tensorflow_source_dir}/tensorflow/cc/framework/gradients_test.cc"
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/call_options_test.cc"
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/tensor_coding_test.cc"
)
if (NOT tensorflow_ENABLE_GPU)
# exclude gpu tests if we are not buildig for gpu
set(tf_test_src_simple_exclude
@ -217,15 +241,6 @@ if (tensorflow_BUILD_CC_TESTS)
)
endif()
# exclude the onces we don't want
set(tf_test_src_simple_exclude
# generally not working
"${tensorflow_source_dir}/tensorflow/cc/client/client_session_test.cc"
"${tensorflow_source_dir}/tensorflow/cc/framework/gradients_test.cc"
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/call_options_test.cc"
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/tensor_coding_test.cc"
)
if (WIN32)
set(tf_test_src_simple_exclude
${tf_test_src_simple_exclude}
@ -235,68 +250,26 @@ if (tensorflow_BUILD_CC_TESTS)
# TODO: test failing
"${tensorflow_source_dir}/tensorflow/core/common_runtime/simple_placer_test.cc"
"${tensorflow_source_dir}/tensorflow/core/debug/debug_gateway_test.cc" # hangs
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/executor_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/quantized_reshape_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/requantization_range_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/requantize_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/restore_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/restore_v2_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/save_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/sparse_reduce_sum_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/restore_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/quantize_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/core/status_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/strings/str_util_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/strings/numbers_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/monitoring/collection_registry_test.cc"
"${tensorflow_source_dir}/tensorflow/core/util/tensor_slice_reader_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/file_system_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/logging_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/env_test.cc"
"${tensorflow_source_dir}/tensorflow/core/ops/math_grad_test.cc"
"${tensorflow_source_dir}/tensorflow/contrib/cudnn_rnn/cudnn_rnn_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/contrib/rnn/ops/gru_ops_test.cc" # status 5
"${tensorflow_source_dir}/tensorflow/contrib/rnn/ops/lstm_ops_test.cc" # status 5
# TODO: not compiling
"${tensorflow_source_dir}/tensorflow/cc/framework/gradient_checker_test.cc"
"${tensorflow_source_dir}/tensorflow/cc/gradients/math_grad_test.cc"
"${tensorflow_source_dir}/tensorflow/cc/gradients/array_grad_test.cc"
"${tensorflow_source_dir}/tensorflow/cc/saved_model/loader_test.cc"
"${tensorflow_source_dir}/tensorflow/cc/training/queue_runner_test.cc"
"${tensorflow_source_dir}/tensorflow/cc/training/coordinator_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/nn_ops_test.cc"
# TODO: not compiling
"${tensorflow_source_dir}/tensorflow/core/kernels/quantization_utils_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/activation_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/batch_norm_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/bias_add_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/concat_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/conv_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/matmul_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/pooling_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/quantize_and_dequantize_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/quantize_down_and_shrink_range_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/parameterized_truncated_normal_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/non_max_suppression_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/fused_batch_norm_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/eigen_backward_spatial_convolutions_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/hexagon/quantized_matmul_op_for_hexagon_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/hexagon/hexagon_graph_transferer_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/adjust_contrast_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/batch_norm_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/cast_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/colorspace_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/control_flow_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/conv_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/debug_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/resize_bilinear_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/resize_nearest_neighbor_op_benchmark_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/spacetobatch_benchmark_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/sparse_add_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/sparse_dense_binary_op_shared_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/sparse_tensor_dense_matmul_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/summary_image_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/summary_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/quantized_activation_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/quantized_bias_add_op_test.cc"
"${tensorflow_source_dir}/tensorflow/core/kernels/quantized_concat_op_test.cc"
@ -312,10 +285,8 @@ if (tensorflow_BUILD_CC_TESTS)
"${tensorflow_source_dir}/tensorflow/core/platform/cloud/retrying_file_system_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/cloud/time_util_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/hadoop/hadoop_file_system_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/port_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/profile_utils/cpu_utils_test.cc"
"${tensorflow_source_dir}/tensorflow/core/common_runtime/direct_session_test.cc"
"${tensorflow_source_dir}/tensorflow/core/common_runtime/direct_session_with_tracking_alloc_test.cc"
"${tensorflow_source_dir}/tensorflow/core/platform/subprocess_test.cc"
"${tensorflow_source_dir}/tensorflow/core/common_runtime/gpu/gpu_allocator_retry_test.cc"
"${tensorflow_source_dir}/tensorflow/core/common_runtime/gpu/gpu_debug_allocator_test.cc"
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/master_test.cc"
@ -325,20 +296,9 @@ if (tensorflow_BUILD_CC_TESTS)
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/rpc/grpc_tensor_coding_test.cc"
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/rpc/rpc_rendezvous_mgr_test.cc"
"${tensorflow_source_dir}/tensorflow/core/distributed_runtime/master_test.cc"
"${tensorflow_source_dir}/tensorflow/core/framework/partial_tensor_shape_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/core/notification_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/gtl/cleanup_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/gtl/edit_distance_test.cc"
"${tensorflow_source_dir}/tensorflow/core/lib/strings/strcat_test.cc"
"${tensorflow_source_dir}/tensorflow/core/ops/array_grad_test.cc"
"${tensorflow_source_dir}/tensorflow/core/ops/nn_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/core/example/example_parser_configuration_test.cc"
"${tensorflow_source_dir}/tensorflow/core/example/feature_util_test.cc"
"${tensorflow_source_dir}/tensorflow/core/util/reporter_test.cc"
"${tensorflow_source_dir}/tensorflow/core/util/memmapped_file_system_test.cc"
"${tensorflow_source_dir}/tensorflow/core/util/sparse_sparse_tensor_test.cc"
"${tensorflow_source_dir}/tensorflow/core/debug/debug_gateway_test.cc"
"${tensorflow_source_dir}/tensorflow/core/debug/debug_io_utils_test.cc"
"${tensorflow_source_dir}/tensorflow/contrib/factorization/kernels/clustering_ops_test.cc"
"${tensorflow_source_dir}/tensorflow/contrib/session_bundle/bundle_shim_test.cc"
"${tensorflow_source_dir}/tensorflow/contrib/session_bundle/bundle_test.cc"
@ -348,7 +308,15 @@ if (tensorflow_BUILD_CC_TESTS)
)
endif()
list(REMOVE_ITEM tf_test_src_simple ${tf_test_src_simple_exclude})
# Tests for saved_model require data, so need to treat them separately.
file(GLOB tf_cc_saved_model_test_srcs
"${tensorflow_source_dir}/tensorflow/cc/saved_model/*_test.cc"
)
list(REMOVE_ITEM tf_test_src_simple
${tf_test_src_simple_exclude}
${tf_cc_saved_model_test_srcs}
)
set(tf_test_lib tf_test_lib)
add_library(${tf_test_lib} STATIC ${tf_src_testlib})
@ -360,6 +328,7 @@ if (tensorflow_BUILD_CC_TESTS)
$<TARGET_OBJECTS:tf_core_cpu>
$<TARGET_OBJECTS:tf_core_framework>
$<TARGET_OBJECTS:tf_core_kernels>
$<TARGET_OBJECTS:tf_cc>
$<TARGET_OBJECTS:tf_cc_framework>
$<TARGET_OBJECTS:tf_cc_ops>
$<TARGET_OBJECTS:tf_core_ops>
@ -375,10 +344,23 @@ if (tensorflow_BUILD_CC_TESTS)
${tensorflow_EXTERNAL_LIBRARIES}
)
# All tests that require no data.
AddTests(
SOURCES ${tf_test_src_simple}
OBJECTS ${tf_obj_test}
LIBS ${tf_test_libs}
DEPENDS googletest
)
# Tests for tensorflow/cc/saved_model.
file(GLOB_RECURSE tf_cc_saved_model_test_data
"${tensorflow_source_dir}/tensorflow/cc/saved_model/testdata/*"
)
AddTests(
SOURCES ${tf_cc_saved_model_test_srcs}
DATA ${tf_cc_saved_model_test_data}
OBJECTS ${tf_obj_test}
LIBS ${tf_test_libs}
)
endif(tensorflow_BUILD_CC_TESTS)

2
tensorflow/contrib/framework/python/ops/prettyprint_ops_test.py
View File

@ -52,7 +52,7 @@ class PrettyPrintOpsTest(tf.test.TestCase):
a = tf.Variable(1.0)
a = tf.contrib.framework.print_op(a)
with self.test_session():
tf.initialize_all_variables().run()
tf.global_variables_initializer().run()
a.eval()
if __name__ == "__main__":

14
tensorflow/contrib/graph_editor/tests/transform_test.py
View File

@ -58,6 +58,20 @@ class TransformTest(tf.test.TestCase):
self.assertEqual(t.name, t_.name)
self.assertEqual(info.original(t_), t)
def test_copy_assert(self):
tf.reset_default_graph()
a = tf.constant(1)
b = tf.constant(1)
eq = tf.equal(a, b)
assert_op = tf.Assert(eq, [a, b])
with tf.control_dependencies([assert_op]):
_ = tf.add(a, b)
sgv = ge.make_view([assert_op, eq.op, a.op, b.op])
copier = ge.Transformer()
copied_sgv, info = copier(sgv, sgv.graph, "", "")
new_assert_op = info.transformed(assert_op)
self.assertIsNotNone(new_assert_op)
def test_transform(self):
transformer = ge.Transformer()
def my_transform_op_handler(info, op):

5
tensorflow/contrib/graph_editor/transform.py
View File

@ -446,10 +446,7 @@ class Transformer(object):
# without any outputs. So the walk is now finalized from those roots.
remaining_ops = [op for op in self._info.sgv.ops
if op not in self._info.transformed_ops]
remaining_roots = [
op for op in remaining_ops
if not op.outputs and not self._info.control_outputs.get(op)
]
remaining_roots = [op for op in remaining_ops if not op.outputs]
for op in remaining_roots:
self._transform_op(op)

8
tensorflow/contrib/layers/python/layers/feature_column_ops_test.py
View File

@ -1209,7 +1209,7 @@ class WeightedSumTest(tf.test.TestCase):
logits, _, _ = tf.contrib.layers.weighted_sum_from_feature_columns(
features, [hashed_sparse], num_outputs=5)
with self.test_session():
tf.initialize_all_variables().run()
tf.global_variables_initializer().run()
self.assertAllEqual(logits.eval().shape, [2, 5])
def testWeightedSparseColumn(self):
@ -1244,7 +1244,7 @@ class WeightedSumTest(tf.test.TestCase):
features, [weighted_ids], num_outputs=5)
with self.test_session():
tf.initialize_all_variables().run()
tf.global_variables_initializer().run()
tf.initialize_all_tables().run()
self.assertAllEqual(logits.eval().shape, [2, 5])
@ -1846,7 +1846,7 @@ class WeightedSumTest(tf.test.TestCase):
[product],
num_outputs=1))
with self.test_session() as sess:
tf.initialize_all_variables().run()
tf.global_variables_initializer().run()
tf.initialize_all_tables().run()
product_weights = column_to_variable[product][0]
sess.run(product_weights.assign([[0.1], [0.2], [0.3], [0.4], [0.5]]))
@ -1862,7 +1862,7 @@ class WeightedSumTest(tf.test.TestCase):
[product],
num_outputs=1))
with self.test_session() as sess:
tf.initialize_all_variables().run()
tf.global_variables_initializer().run()
tf.initialize_all_tables().run()
product_weights = column_to_variable[product][0]
sess.run(product_weights.assign([[0.1], [0.2], [0.3], [0.4], [0.5]]))

12
tensorflow/contrib/layers/python/layers/layers.py
View File

@ -171,7 +171,10 @@ def _fused_batch_norm(
`batch_size`. The normalization is over all but the last dimension if
`data_format` is `NHWC` and the second dimension if `data_format` is
`NCHW`.
decay: decay for the moving average.
decay: decay for the moving average. Reasonable values for `decay` are close
to 1.0, typically in the multiple-nines range: 0.999, 0.99, 0.9, etc. Lower
`decay` value (recommend trying `decay`=0.9) if model experiences reasonably
good training performance but poor validation and/or test performance.
center: If True, subtract `beta`. If False, `beta` is ignored.
scale: If True, multiply by `gamma`. If False, `gamma` is
not used. When the next layer is linear (also e.g. `nn.relu`), this can be
@ -396,7 +399,10 @@ def batch_norm(
`batch_size`. The normalization is over all but the last dimension if
`data_format` is `NHWC` and the second dimension if `data_format` is
`NCHW`.
decay: decay for the moving average.
decay: decay for the moving average. Reasonable values for `decay` are close
to 1.0, typically in the multiple-nines range: 0.999, 0.99, 0.9, etc. Lower
`decay` value (recommend trying `decay`=0.9) if model experiences reasonably
good training performance but poor validation and/or test performance.
center: If True, subtract `beta`. If False, `beta` is ignored.
scale: If True, multiply by `gamma`. If False, `gamma` is
not used. When the next layer is linear (also e.g. `nn.relu`), this can be
@ -1369,7 +1375,7 @@ def fully_connected(inputs,
Raises:
ValueError: if x has rank less than 2 or if its last dimension is not set.
"""
if not (isinstance(num_outputs, int) or isinstance(num_outputs, long)):
if not (isinstance(num_outputs, six.integer_types)):
raise ValueError('num_outputs should be int or long, got %s.', num_outputs)
layer_variable_getter = _build_variable_getter({'bias': 'biases'})

4
tensorflow/contrib/layers/python/layers/optimizers_test.py
View File

@ -195,7 +195,7 @@ class OptimizersTest(tf.test.TestCase):
self.assertAlmostEqual(var_value, 9.8916, 4)
self.assertEqual(global_step_value, 1)
var_count = 0
for var in tf.all_variables():
for var in tf.global_variables():
if var.name.startswith("OptimizeLoss/AdaptiveMaxNorm"):
var_count += 1
self.assertEqual(2, var_count)
@ -366,7 +366,7 @@ class AdaptiveClipping(tf.test.TestCase):
decay=0.5)(grads_and_vars)
var_dict = {}
for var in tf.all_variables():
for var in tf.global_variables():
if var.name.startswith("AdaptiveMaxNorm"):
var_dict[var.name.split(":")[0]] = var
self.assertEqual(2, len(var_dict))

16
tensorflow/contrib/learn/python/learn/datasets/base.py
View File

@ -186,8 +186,8 @@ def _is_retriable(e):
@retry(initial_delay=1.0, max_delay=16.0, is_retriable=_is_retriable)
def urlretrieve_with_retry(url, filename):
urllib.request.urlretrieve(url, filename)
def urlretrieve_with_retry(url, filename=None):
return urllib.request.urlretrieve(url, filename)
def maybe_download(filename, work_directory, source_url):
@ -205,11 +205,9 @@ def maybe_download(filename, work_directory, source_url):
gfile.MakeDirs(work_directory)
filepath = os.path.join(work_directory, filename)
if not gfile.Exists(filepath):
with tempfile.NamedTemporaryFile() as tmpfile:
temp_file_name = tmpfile.name
urlretrieve_with_retry(source_url, temp_file_name)
gfile.Copy(temp_file_name, filepath)
with gfile.GFile(filepath) as f:
size = f.size()
print('Successfully downloaded', filename, size, 'bytes.')
temp_file_name, _ = urlretrieve_with_retry(source_url)
gfile.Copy(temp_file_name, filepath)
with gfile.GFile(filepath) as f:
size = f.size()
print('Successfully downloaded', filename, size, 'bytes.')
return filepath

70
tensorflow/contrib/learn/python/learn/estimators/estimator.py
View File

@ -330,8 +330,8 @@ class BaseEstimator(
# Features and labels TensorSignature objects.
# TODO(wicke): Rename these to something more descriptive
self._features_info = None
self._labels_info = None
self._features_info = {}
self._labels_info = {}
self._graph = None
@ -641,28 +641,29 @@ class BaseEstimator(
return tensor_signature.create_example_parser_from_signatures(
self._features_info, examples_batch)
def _check_inputs(self, features, labels):
if self._features_info is not None:
logging.debug('Given features: %s, required signatures: %s.',
str(features), str(self._features_info))
if not tensor_signature.tensors_compatible(features, self._features_info):
raise ValueError('Features are incompatible with given information. '
def _check_inputs(self, features, labels, mode):
if mode in self._features_info:
logging.debug('Given features for mode %s: %s, required signatures: %s.',
mode, str(features), str(self._features_info[mode]))
if not tensor_signature.tensors_compatible(features, self._features_info[mode]):
raise ValueError('Features for mode %s are incompatible with given information. '
'Given features: %s, required signatures: %s.' %
(str(features), str(self._features_info)))
(mode, str(features), str(self._features_info[mode])))
else:
self._features_info = tensor_signature.create_signatures(features)
logging.debug('Setting feature info to %s.', str(self._features_info))
self._features_info[mode] = tensor_signature.create_signatures(features)
logging.debug('Setting feature info for mode %s to %s.', mode, str(self._features_info[mode]))
if labels is not None:
if self._labels_info is not None:
if mode in self._labels_info:
logging.debug('Given labels: %s, required signatures: %s.',
str(labels), str(self._labels_info))
if not tensor_signature.tensors_compatible(labels, self._labels_info):
raise ValueError('Labels are incompatible with given information. '
if not tensor_signature.tensors_compatible(labels, self._labels_info[mode]):
raise ValueError('Labels for mode %s are incompatible with given information. '
'Given labels: %s, required signatures: %s.' %
(str(labels), str(self._labels_info)))
(mode, str(labels), str(self._labels_info[mode])))
else:
self._labels_info = tensor_signature.create_signatures(labels)
logging.debug('Setting labels info to %s', str(self._labels_info))
self._labels_info[mode] = tensor_signature.create_signatures(labels)
logging.debug('Setting labels info for mode %s to %s', mode, str(self._labels_info[mode]))
def _train_model(self,
input_fn,
@ -699,8 +700,7 @@ class BaseEstimator(
random_seed.set_random_seed(self._config.tf_random_seed)
global_step = contrib_framework.create_global_step(g)
features, labels = input_fn()
self._check_inputs(features, labels)
self._check_inputs(features, labels, model_fn_lib.ModeKeys.TRAIN)
# The default return type of _get_train_ops is ModelFnOps. But there are
# some subclasses of tf.contrib.learn.Estimator which override this
# method and use the legacy signature, namely _get_train_ops returns a
@ -800,8 +800,7 @@ class BaseEstimator(
random_seed.set_random_seed(self._config.tf_random_seed)
global_step = contrib_framework.create_global_step(g)
features, labels = input_fn()
self._check_inputs(features, labels)
self._check_inputs(features, labels, model_fn_lib.ModeKeys.EVAL)
# The default return type of _get_eval_ops is ModelFnOps. But there are
# some subclasses of tf.contrib.learn.Estimator which override this
# method and use the legacy signature, namely _get_eval_ops returns an
@ -835,6 +834,29 @@ class BaseEstimator(
return result[0]
return result
def _set_infer_mode_feature_signature(self, features):
for mode in list(self._features_info.keys()):
if tensor_signature.tensors_compatible(features, self._features_info[mode]):
self._features_info[model_fn_lib.ModeKeys.INFER] = self._features_info[mode]
if mode in self._labels_info:
self._labels_info[model_fn_lib.ModeKeys.INFER] = (
self._labels_info[mode])
else:
self._labels_info[model_fn_lib.ModeKeys.INFER] = None
break
if model_fn_lib.ModeKeys.INFER not in self._features_info:
logging.warning('Features for mode %s are incompatible with neither train mode nor eval mode.'
' Given features: %s' % (model_fn_lib.ModeKeys.INFER, str(features)))
for mode in list(self._features_info.keys()):
logging.warning('Whereas %s mode signatures: %s' % (mode, str(self._features_info[mode])))
self._check_inputs(features, None, model_fn_lib.ModeKeys.INFER)
if model_fn_lib.ModeKeys.TRAIN in self._labels_info:
logging.warning('Setting labels info for mode infer equal to that of labels info for train mode')
self._labels_info[model_fn_lib.ModeKeys.INFER] = self._labels_info[model_fn_lib.ModeKeys.TRAIN]
else:
self._labels_info[model_fn_lib.ModeKeys.INFER] = {}
def _infer_model(
self, input_fn, feed_fn=None, outputs=None, as_iterable=True):
# Check that model has been trained.
@ -1134,8 +1156,10 @@ class Estimator(BaseEstimator):
Returns:
`ModelFnOps` object.
"""
self._set_infer_mode_feature_signature(features)
labels = tensor_signature.create_placeholders_from_signatures(
self._labels_info)
self._labels_info[model_fn_lib.ModeKeys.INFER])
return self._call_model_fn(features, labels, model_fn_lib.ModeKeys.INFER)
@experimental
@ -1239,7 +1263,7 @@ class Estimator(BaseEstimator):
return export_dir
# For time of deprecation x,y from Estimator allow direct access.
# For time of deprecation x,y from Estimator allow direct access
# pylint: disable=protected-access
class SKCompat(sklearn.BaseEstimator):
"""Scikit learn wrapper for TensorFlow Learn Estimator."""

86
tensorflow/contrib/learn/python/learn/estimators/estimator_test.py
View File

@ -91,7 +91,18 @@ def boston_eval_fn():
0)
def extract(data, key):
if isinstance(data, dict):
assert key in data
return data[key]
else:
return data
def linear_model_params_fn(features, labels, mode, params):
features = extract(features, 'input')
labels = extract(labels, 'labels')
assert mode in (
tf.contrib.learn.ModeKeys.TRAIN,
tf.contrib.learn.ModeKeys.EVAL,
@ -106,6 +117,8 @@ def linear_model_params_fn(features, labels, mode, params):
def linear_model_fn(features, labels, mode):
features = extract(features, 'input')
labels = extract(labels, 'labels')
assert mode in (
tf.contrib.learn.ModeKeys.TRAIN,
tf.contrib.learn.ModeKeys.EVAL,
@ -140,8 +153,8 @@ def linear_model_fn_with_model_fn_ops(features, labels, mode):
def logistic_model_no_mode_fn(features, labels):
if isinstance(labels, dict):
labels = labels['labels']
features = extract(features, 'input')
labels = extract(labels, 'labels')
labels = tf.one_hot(labels, 3, 1, 0)
prediction, loss = (
tf.contrib.learn.models.logistic_regression_zero_init(features, labels)
@ -346,6 +359,34 @@ class EstimatorTest(tf.test.TestCase):
with self.assertRaises(tf.contrib.learn.NotFittedError):
est.predict(x=boston.data)
def testContinueTrainingDictionaryInput(self):
boston = tf.contrib.learn.datasets.load_boston()
output_dir = tempfile.mkdtemp()
est = tf.contrib.learn.Estimator(model_fn=linear_model_fn,
model_dir=output_dir)
boston_input = {'input': boston.data}
float64_target = {'labels': boston.target.astype(np.float64)}
est.fit(x=boston_input, y=float64_target, steps=50)
scores = est.evaluate(
x=boston_input,
y=float64_target,
metrics={'MSE': tf.contrib.metrics.streaming_mean_squared_error})
del est
# Create another estimator object with the same output dir.
est2 = tf.contrib.learn.Estimator(model_fn=linear_model_fn,
model_dir=output_dir)
# Check we can evaluate and predict.
scores2 = est2.evaluate(
x=boston_input,
y=float64_target,
metrics={'MSE': tf.contrib.metrics.streaming_mean_squared_error})
self.assertAllClose(scores2['MSE'],
scores['MSE'])
predictions = np.array(list(est2.predict(x=boston_input)))
other_score = _sklearn.mean_squared_error(predictions, float64_target['labels'])
self.assertAllClose(other_score, scores['MSE'])
def testContinueTraining(self):
boston = tf.contrib.learn.datasets.load_boston()
output_dir = tempfile.mkdtemp()
@ -405,6 +446,22 @@ class EstimatorTest(tf.test.TestCase):
self.assertTrue('global_step' in scores)
self.assertEqual(100, scores['global_step'])
def testBostonAllDictionaryInput(self):
boston = tf.contrib.learn.datasets.load_boston()
est = tf.contrib.learn.Estimator(model_fn=linear_model_fn)
boston_input = {'input': boston.data}
float64_target = {'labels': boston.target.astype(np.float64)}
est.fit(x=boston_input, y=float64_target, steps=100)
scores = est.evaluate(
x=boston_input,
y=float64_target,
metrics={'MSE': tf.contrib.metrics.streaming_mean_squared_error})
predictions = np.array(list(est.predict(x=boston_input)))
other_score = _sklearn.mean_squared_error(predictions, boston.target)
self.assertAllClose(other_score, scores['MSE'])
self.assertTrue('global_step' in scores)
self.assertEqual(scores['global_step'], 100)
def testIrisAll(self):
iris = tf.contrib.learn.datasets.load_iris()
est = tf.contrib.learn.SKCompat(
@ -428,6 +485,31 @@ class EstimatorTest(tf.test.TestCase):
self.assertTrue('global_step' in scores)
self.assertEqual(100, scores['global_step'])
def testIrisAllDictionaryInput(self):
iris = tf.contrib.learn.datasets.load_iris()
est = tf.contrib.learn.Estimator(model_fn=logistic_model_no_mode_fn)
iris_data = {'input': iris.data}
iris_target = {'labels': iris.target}
est.fit(iris_data, iris_target, steps=100)
scores = est.evaluate(
x=iris_data,
y=iris_target,
metrics={('accuracy', 'class'): tf.contrib.metrics.streaming_accuracy})
predictions = list(est.predict(x=iris_data))
predictions_class = list(est.predict(x=iris_data, outputs=['class']))
self.assertEqual(len(predictions), iris.target.shape[0])
classes_batch = np.array([p['class'] for p in predictions])
self.assertAllClose(
classes_batch,
np.array([p['class'] for p in predictions_class]))
self.assertAllClose(
classes_batch,
np.argmax(np.array([p['prob'] for p in predictions]), axis=1))
other_score = _sklearn.accuracy_score(iris.target, classes_batch)
self.assertAllClose(other_score, scores['accuracy'])
self.assertTrue('global_step' in scores)
self.assertEqual(scores['global_step'], 100)
def testIrisInputFn(self):
iris = tf.contrib.learn.datasets.load_iris()
est = tf.contrib.learn.Estimator(model_fn=logistic_model_no_mode_fn)

3
tensorflow/contrib/learn/python/learn/estimators/linear.py
View File

@ -660,9 +660,10 @@ class LinearRegressor(evaluable.Evaluable, trainable.Trainable):
"""
self._feature_columns = feature_columns
assert self._feature_columns
self._optimizer = _get_default_optimizer(feature_columns)
if optimizer:
self._optimizer = _get_optimizer(optimizer)
else:
self._optimizer = _get_default_optimizer(feature_columns)
chief_hook = None
if (isinstance(optimizer, sdca_optimizer.SDCAOptimizer) and

10
tensorflow/contrib/learn/python/learn/evaluable.py
View File

@ -51,12 +51,14 @@ class Evaluable(object):
for which this evaluation was performed.
Args:
x: Matrix of shape [n_samples, n_features...] containing the input samples
for fitting the model. Can be iterator that returns arrays of features.
If set, `input_fn` must be `None`.
x: Matrix of shape [n_samples, n_features...] or dictionary of many matrices
containing the input samples for fitting the model. Can be iterator that returns
arrays of features or dictionary of array of features. If set, `input_fn` must
be `None`.
y: Vector or matrix [n_samples] or [n_samples, n_outputs] containing the
label values (class labels in classification, real numbers in
regression). Can be iterator that returns array of labels. If set,
regression) or dictionary of multiple vectors/matrices. Can be iterator
that returns array of targets or dictionary of array of targets. If set,
`input_fn` must be `None`. Note: For classification, label values must
be integers representing the class index (i.e. values from 0 to
n_classes-1).

2
tensorflow/contrib/learn/python/learn/graph_actions.py
View File

@ -299,10 +299,10 @@ def _monitored_train(graph,
while not super_sess.should_stop():
_, loss = super_sess.run([train_op, loss_op], feed_fn() if feed_fn else
None)
summary_io.SummaryWriterCache.clear()
return loss
# TODO(ispir): Deprecate train in favor of supervised_train
def train(graph,
output_dir,

497
tensorflow/contrib/learn/python/learn/learn_io/data_feeder.py
View File

@ -36,27 +36,49 @@ from tensorflow.python.platform import tf_logging as logging
# pylint: disable=g-multiple-import,g-bad-import-order
from .pandas_io import HAS_PANDAS, extract_pandas_data, extract_pandas_matrix, extract_pandas_labels
from .dask_io import HAS_DASK, extract_dask_data, extract_dask_labels
# pylint: enable=g-multiple-import,g-bad-import-order
def _get_in_out_shape(x_shape, y_shape, n_classes, batch_size=None):
"""Returns shape for input and output of the data feeder."""
x_is_dict, y_is_dict = isinstance(x_shape, dict), y_shape is not None and isinstance(y_shape, dict)
if y_is_dict and n_classes is not None:
assert (isinstance(n_classes, dict))
if batch_size is None:
batch_size = x_shape[0]
batch_size = list(x_shape.values())[0][0] if x_is_dict else x_shape[0]
elif batch_size <= 0:
raise ValueError('Invalid batch_size %d.' % batch_size)
x_shape = list(x_shape[1:]) if len(x_shape) > 1 else [1]
input_shape = [batch_size] + x_shape
if x_is_dict:
input_shape = {}
for k, v in list(x_shape.items()):
input_shape[k] = [batch_size] + (list(v[1:]) if len(v) > 1 else [1])
else:
x_shape = list(x_shape[1:]) if len(x_shape) > 1 else [1]
input_shape = [batch_size] + x_shape
if y_shape is None:
return input_shape, None, batch_size
y_shape = list(y_shape[1:]) if len(y_shape) > 1 else []
# Skip first dimension if it is 1.
if y_shape and y_shape[0] == 1:
y_shape = y_shape[1:]
if n_classes is not None and n_classes > 1:
output_shape = [batch_size] + y_shape + [n_classes]
def out_el_shape(out_shape, num_classes):
out_shape = list(out_shape[1:]) if len(out_shape) > 1 else []
# Skip first dimension if it is 1.
if out_shape and out_shape[0] == 1:
out_shape = out_shape[1:]
if num_classes is not None and num_classes > 1:
return [batch_size] + out_shape + [num_classes]
else:
return [batch_size] + out_shape
if not y_is_dict:
output_shape = out_el_shape(y_shape, n_classes)
else:
output_shape = [batch_size] + y_shape
output_shape = dict([(k, out_el_shape(v, n_classes[k] if n_classes is not None and k in n_classes else None))
for k, v in list(y_shape.items())])
return input_shape, output_shape, batch_size
@ -78,15 +100,18 @@ def _is_iterable(x):
def setup_train_data_feeder(
x, y, n_classes, batch_size=None, shuffle=True, epochs=None):
x, y, n_classes, batch_size=None, shuffle=True, epochs=None):
"""Create data feeder, to sample inputs from dataset.
If `x` and `y` are iterators, use `StreamingDataFeeder`.
Args:
x: numpy, pandas or Dask matrix or iterable.
y: numpy, pandas or Dask array or iterable.
n_classes: number of classes.
x: numpy, pandas or Dask matrix or dictionary of aforementioned. Also
supports iterables.
y: numpy, pandas or Dask array or dictionary of aforementioned. Also supports
iterables.
n_classes: number of classes. Must be None or same type as y. In case, `y` is `dict`
(or iterable which returns dict) such that `n_classes[key] = n_classes for y[key]`
batch_size: size to split data into parts. Must be >= 1.
shuffle: Whether to shuffle the inputs.
epochs: Number of epochs to run.
@ -102,7 +127,7 @@ def setup_train_data_feeder(
# pylint: disable=g-import-not-at-top
import dask.dataframe as dd
if (isinstance(x, (dd.Series, dd.DataFrame)) and
(y is None or isinstance(y, (dd.Series, dd.DataFrame)))):
(y is None or isinstance(y, (dd.Series, dd.DataFrame)))):
data_feeder_cls = DaskDataFeeder
else:
data_feeder_cls = DataFeeder
@ -115,31 +140,54 @@ def setup_train_data_feeder(
'streaming learning to work.')
return StreamingDataFeeder(x, y, n_classes, batch_size)
return data_feeder_cls(
x, y, n_classes, batch_size, shuffle=shuffle, epochs=epochs)
x, y, n_classes, batch_size, shuffle=shuffle, epochs=epochs)
def _batch_data(x, batch_size=None):
if (batch_size is not None) and (batch_size <= 0):
raise ValueError('Invalid batch_size %d.' % batch_size)
chunk = []
x_first_el = six.next(x)
x = itertools.chain([x_first_el], x)
chunk = dict([(k, []) for k in list(x_first_el.keys())]) if isinstance(x_first_el, dict) else []
chunk_filled = False
for data in x:
chunk.append(data)
if (batch_size is not None) and (len(chunk) >= batch_size):
yield np.matrix(chunk)
chunk = []
yield np.matrix(chunk)
if isinstance(data, dict):
for k, v in list(data.items()):
chunk[k].append(v)
if (batch_size is not None) and (len(chunk[k]) >= batch_size):
chunk[k] = np.matrix(chunk[k])
chunk_filled = True
if chunk_filled:
yield chunk
chunk = dict([(k, []) for k in list(x_first_el.keys())]) if isinstance(x_first_el, dict) else []
chunk_filled = False
else:
chunk.append(data)
if (batch_size is not None) and (len(chunk) >= batch_size):
yield np.matrix(chunk)
chunk = []
if isinstance(x_first_el, dict):
for k, v in list(data.items()):
chunk[k] = np.matrix(chunk[k])
yield chunk
else:
yield np.matrix(chunk)
def setup_predict_data_feeder(x, batch_size=None):
"""Returns an iterable for feeding into predict step.
Args:
x: numpy, pandas, Dask array or iterable.
batch_size: Size of batches to split data into.
If `None`, returns one batch of full size.
x: numpy, pandas, Dask array or dictionary of aforementioned. Also supports
iterable.
batch_size: Size of batches to split data into. If `None`, returns one
batch of full size.
Returns:
List or iterator of parts of data to predict on.
List or iterator (or dictionary thereof) of parts of data to predict on.
Raises:
ValueError: if `batch_size` <= 0.
@ -211,7 +259,7 @@ def _access(data, iloc):
def _check_dtype(dtype):
if dtypes.as_dtype(dtype) == dtypes.float64:
logging.warn(
'float64 is not supported by many models, consider casting to float32.')
'float64 is not supported by many models, consider casting to float32.')
return dtype
@ -219,63 +267,85 @@ class DataFeeder(object):
"""Data feeder is an example class to sample data for TF trainer."""
def __init__(
self, x, y, n_classes, batch_size=None, shuffle=True, random_state=None,
epochs=None):
self, x, y, n_classes, batch_size=None, shuffle=True, random_state=None,
epochs=None):
"""Initializes a DataFeeder instance.
Args:
x: Feature Nd numpy matrix of shape `[n_samples, n_features, ...]`.
y: Label vector, either floats for regression or class id for
classification. If matrix, will consider as a sequence
of labels. Can be `None` for unsupervised setting.
x: One feature sample which can either Nd numpy matrix of shape
`[n_samples, n_features, ...]` or dictionary of Nd numpy matrix.
y: label vector, either floats for regression or class id for
classification. If matrix, will consider as a sequence of labels.
Can be `None` for unsupervised setting. Also supports dictionary of
labels.
n_classes: Number of classes, 0 and 1 are considered regression, `None`
will pass through the input labels without one-hot conversion.
batch_size: Mini-batch size to accumulate.
will pass through the input labels without one-hot conversion. Also, if
`y` is `dict`, then `n_classes` must be `dict` such that
`n_classes[key] = n_classes for label y[key]`, `None` otherwise.
batch_size: Mini-batch size to accumulate samples in one mini batch.
shuffle: Whether to shuffle `x`.
random_state: Numpy `RandomState` object to reproduce sampling.
epochs: Number of times to iterate over input data before raising
`StopIteration` exception.
Attributes:
x: Input features.
y: Input label.
x: Input features (ndarray or dictionary of ndarrays).
y: Input label (ndarray or dictionary of ndarrays).
n_classes: Number of classes (if `None`, pass through indices without
one-hot conversion).
batch_size: Mini-batch size to accumulate.
input_shape: Shape of the input.
output_shape: Shape of the output.
input_dtype: DType of input.
output_dtype: DType of output.
input_shape: Shape of the input (or dictionary of shapes).
output_shape: Shape of the output (or dictionary of shapes).
input_dtype: DType of input (or dictionary of shapes).
output_dtype: DType of output (or dictionary of shapes.
"""
self._x = check_array(x, dtype=x.dtype)
# self.n_classes is None means we're passing in raw label indices.
y_dtype = (
np.int64 if n_classes is not None and n_classes > 1 else np.float32)
x_is_dict, y_is_dict = isinstance(x, dict), y is not None and isinstance(y, dict)
if isinstance(y, list):
y = np.array(y)
self._x = dict([(k, check_array(v, v.dtype)) for k, v in list(x.items())]) if x_is_dict else check_array(x, x.dtype)
self._y = None if y is None else \
dict([(k, check_array(v, v.dtype)) for k, v in list(y.items())]) if x_is_dict else check_array(y, y.dtype)
# self.n_classes is not None means we're converting raw target indices to one-hot.
if n_classes is not None:
self._y = (None if y is None else check_array(y, dtype=y_dtype))
elif isinstance(y, list):
self._y = np.array(y)
else:
self._y = y
if not y_is_dict:
y_dtype = (np.int64 if n_classes is not None and n_classes > 1 else np.float32)
self._y = (None if y is None else check_array(y, dtype=y_dtype))
self.n_classes = n_classes
self.max_epochs = epochs
x_shape = dict([(k, v.shape) for k, v in list(self._x.items())]) if x_is_dict else self._x.shape
y_shape = dict(
[(k, v.shape) for k, v in list(self._y.items())]) if y_is_dict else None if y is None else self._y.shape
self.input_shape, self.output_shape, self._batch_size = _get_in_out_shape(
self._x.shape, None if self._y is None else self._y.shape, n_classes,
batch_size)
x_shape, y_shape, n_classes, batch_size)
# Input dtype matches dtype of x.
self._input_dtype = _check_dtype(self._x.dtype)
# self.n_classes is None means we're passing in raw label indices
if n_classes is not None or self._y is None:
self._output_dtype = np.float32
else:
self._output_dtype = _check_dtype(self._y.dtype)
self._input_dtype = dict([(k, _check_dtype(v.dtype)) for k, v in list(self._x.items())]) if x_is_dict \
else _check_dtype(self._x.dtype)
# note: self._output_dtype = np.float32 when y is None
self._output_dtype = dict([(k, _check_dtype(v.dtype)) for k, v in list(self._y.items())]) if y_is_dict \
else _check_dtype(self._y.dtype) if y is not None else np.float32
# self.n_classes is None means we're passing in raw target indices
if n_classes is not None and y_is_dict:
for key in list(n_classes.keys()):
if key in self._output_dtype:
self._output_dtype[key] = np.float32
self._shuffle = shuffle
self.random_state = np.random.RandomState(
42) if random_state is None else random_state
42) if random_state is None else random_state
num_samples = list(self._x.values())[0].shape[0] if x_is_dict else self._x.shape[0]
if self._shuffle:
self.indices = self.random_state.permutation(self._x.shape[0])
self.indices = self.random_state.permutation(num_samples)
else:
self.indices = np.array(range(self._x.shape[0]))
self.indices = np.array(range(num_samples))
self.offset = 0
self.epoch = 0
self._epoch_placeholder = None
@ -320,19 +390,27 @@ class DataFeeder(object):
Returns:
Two placeholders for inputs and outputs.
"""
input_shape = [None] + self.input_shape[1:]
self._input_placeholder = array_ops.placeholder(
dtypes.as_dtype(self._input_dtype),
input_shape,
name='input')
if self.output_shape is None:
self._output_placeholder = None
else:
output_shape = [None] + self.output_shape[1:]
self._output_placeholder = array_ops.placeholder(
dtypes.as_dtype(self._output_dtype),
output_shape,
name='output')
def get_placeholder(shape, dtype, name_prepend):
if shape is None:
return None
if isinstance(shape, dict):
placeholder = {}
for key in list(shape.keys()):
placeholder[key] = array_ops.placeholder(
dtypes.as_dtype(dtype[key]),
[None] + shape[key][1:],
name=name_prepend + '_' + key
)
else:
placeholder = array_ops.placeholder(
dtypes.as_dtype(dtype),
[None] + shape[1:],
name=name_prepend)
return placeholder
self._input_placeholder = get_placeholder(self.input_shape, self._input_dtype, 'input')
self._output_placeholder = get_placeholder(self.output_shape, self._output_dtype, 'output')
return self._input_placeholder, self._output_placeholder
def set_placeholders(self, input_placeholder, output_placeholder):
@ -342,21 +420,21 @@ class DataFeeder(object):
input_placeholder: Placeholder for `x` variable. Should match shape
of the examples in the x dataset.
output_placeholder: Placeholder for `y` variable. Should match
shape of the examples in the y dataset. Can be None.
shape of the examples in the y dataset. Can be `None`.
"""
self._input_placeholder = input_placeholder
self._output_placeholder = output_placeholder
def get_feed_params(self):
"""Function returns a dict with data feed params while training.
"""Function returns a `dict` with data feed params while training.
Returns:
A dict with data feed params while training.
A `dict` with data feed params while training.
"""
return {
'epoch': self.epoch,
'offset': self.offset,
'batch_size': self._batch_size
'epoch': self.epoch,
'offset': self.offset,
'batch_size': self._batch_size
}
def get_feed_dict_fn(self):
@ -364,8 +442,35 @@ class DataFeeder(object):
Returns:
A function that when called samples a random subset of batch size
from x and y.
from `x` and `y`.
"""
x_is_dict, y_is_dict = isinstance(self._x, dict), self._y is not None and isinstance(self._y, dict)
# Assign input features from random indices.
def extract(data, indices):
return (np.array(_access(data, indices)).reshape((indices.shape[0], 1))
if len(data.shape) == 1 else _access(data, indices))
# assign labels from random indices
def assign_label(data, shape, dtype, n_classes, indices):
shape[0] = indices.shape[0]
out = np.zeros(shape, dtype=dtype)
for i in xrange(out.shape[0]):
sample = indices[i]
# self.n_classes is None means we're passing in raw target indices
if n_classes is None:
out[i] = _access(data, sample)
else:
if n_classes > 1:
if len(shape) == 2:
out.itemset((i, int(_access(data, sample))), 1.0)
else:
for idx, value in enumerate(_access(data, sample)):
out.itemset(tuple([i, idx, value]), 1.0)
else:
out[i] = _access(data, sample)
return out
def _feed_dict_fn():
"""Function that samples data into given placeholders."""
if self.max_epochs is not None and self.epoch + 1 > self.max_epochs:
@ -376,20 +481,19 @@ class DataFeeder(object):
feed_dict[self._epoch_placeholder.name] = [self.epoch]
# Take next batch of indices.
end = min(self._x.shape[0], self.offset + self._batch_size)
x_len = list(self._x.values())[0].shape[0] if x_is_dict else self._x.shape[0]
end = min(x_len, self.offset + self._batch_size)
batch_indices = self.indices[self.offset:end]
# Assign input features from random indices.
inp = (
np.array(_access(self._x, batch_indices)).reshape(
(batch_indices.shape[0], 1))
if len(self._x.shape) == 1 else _access(self._x, batch_indices))
feed_dict[self._input_placeholder.name] = inp
# adding input placeholder
feed_dict.update(
dict([(self._input_placeholder[k].name, extract(v, batch_indices)) for k, v in list(self._x.items())])
if x_is_dict else {self._input_placeholder.name: extract(self._x, batch_indices)})
# move offset and reset it if necessary
self.offset += self._batch_size
if self.offset >= self._x.shape[0]:
self.indices = self.random_state.permutation(self._x.shape[0])
if self.offset >= x_len:
self.indices = self.random_state.permutation(x_len) if self._shuffle else np.array(range(x_len))
self.offset = 0
self.epoch += 1
@ -397,24 +501,18 @@ class DataFeeder(object):
if self._output_placeholder is None:
return feed_dict
# assign labels from random indices
self.output_shape[0] = batch_indices.shape[0]
out = np.zeros(self.output_shape, dtype=self._output_dtype)
for i in xrange(out.shape[0]):
sample = batch_indices[i]
# self.n_classes is None means we're passing in raw label indices
if self.n_classes is None:
out[i] = _access(self._y, sample)
else:
if self.n_classes > 1:
if len(self.output_shape) == 2:
out.itemset((i, int(_access(self._y, sample))), 1.0)
else:
for idx, value in enumerate(_access(self._y, sample)):
out.itemset(tuple([i, idx, value]), 1.0)
else:
out[i] = _access(self._y, sample)
feed_dict[self._output_placeholder.name] = out
# adding output placeholders
if y_is_dict:
for k, v in list(self._y.items()):
n_classes = (
self.n_classes[k] if k in self.n_classes else None) if self.n_classes is not None else None
shape, dtype = self.output_shape[k], self._output_dtype[k]
feed_dict.update(
{self._output_placeholder[k].name: assign_label(v, shape, dtype, n_classes, batch_indices)})
else:
shape, dtype, n_classes = self.output_shape, self._output_dtype, self.n_classes
feed_dict.update(
{self._output_placeholder.name: assign_label(self._y, shape, dtype, n_classes, batch_indices)})
return feed_dict
@ -433,21 +531,29 @@ class StreamingDataFeeder(DataFeeder):
"""Initializes a StreamingDataFeeder instance.
Args:
x: iterator that returns for each element, returns features.
y: iterator that returns for each element, returns 1 or many classes /
regression values.
n_classes: indicator of how many classes the label has.
batch_size: Mini batch size to accumulate.
x: iterator each element of which returns one feature sample. Sample can
be a Nd numpy matrix or dictionary of Nd numpy matrices.
y: iterator each element of which returns one label sample. Sample can be
a Nd numpy matrix or dictionary of Nd numpy matrices with 1 or many
classes regression values.
n_classes: indicator of how many classes the corresponding label sample
has for the purposes of one-hot conversion of label. In case where `y`
is a dictionary, `n_classes` must be dictionary (with same keys as `y`)
of how many classes there are in each label in `y`. If key is
present in `y` and missing in `n_classes`, the value is assumed `None`
and no one-hot conversion will be applied to the label with that key.
batch_size: Mini batch size to accumulate samples in one batch. If set
`None`, then assumes that iterator to return already batched element.
Attributes:
x: input features.
y: input label.
x: input features (or dictionary of input features).
y: input label (or dictionary of output features).
n_classes: number of classes.
batch_size: mini batch size to accumulate.
input_shape: shape of the input.
output_shape: shape of the output.
input_dtype: dtype of input.
output_dtype: dtype of output.
input_shape: shape of the input (can be dictionary depending on `x`).
output_shape: shape of the output (can be dictionary depending on `y`).
input_dtype: dtype of input (can be dictionary depending on `x`).
output_dtype: dtype of output (can be dictionary depending on `y`).
"""
# pylint: disable=invalid-name,super-init-not-called
x_first_el = six.next(x)
@ -459,25 +565,48 @@ class StreamingDataFeeder(DataFeeder):
y_first_el = None
self._y = None
self.n_classes = n_classes
x_first_el = ops.convert_to_tensor(x_first_el)
y_first_el = ops.convert_to_tensor(y_first_el) if y is not None else None
self.input_shape, self.output_shape, self._batch_size = _get_in_out_shape(
[1] + list(x_first_el.get_shape()),
[1] + list(y_first_el.get_shape()) if y is not None else None,
n_classes,
batch_size)
self._input_dtype = _check_dtype(x_first_el.dtype).as_numpy_dtype
x_is_dict, y_is_dict = isinstance(x_first_el, dict), y is not None and isinstance(y_first_el, dict)
if y_is_dict and n_classes is not None:
assert (isinstance(n_classes, dict))
# extract shapes for first_elements
x_first_el_shape = dict([(k, [1] + list(v.shape)) for k, v in list(x_first_el.items())]) if x_is_dict \
else [1] + list(x_first_el.shape)
y_first_el_shape = dict([(k, [1] + list(v.shape)) for k, v in list(y_first_el.items())]) if y_is_dict \
else ([1] + list(y_first_el[0].shape if isinstance(y_first_el, list) else y_first_el.shape)
if y is not None else None)
self.input_shape, self.output_shape, self._batch_size = _get_in_out_shape(x_first_el_shape, y_first_el_shape,
n_classes, batch_size)
# Input dtype of x_first_el.
self._input_dtype = dict([(k, _check_dtype(v.dtype)) for k, v in list(x_first_el.items())]) if x_is_dict \
else _check_dtype(x_first_el.dtype)
# Output dtype of y_first_el.
def check_y_dtype(el):
if isinstance(el, list) or isinstance(el, np.ndarray):
if isinstance(el, np.ndarray) and el.ndim == 0:
return el.dtype
else:
return _check_dtype(np.dtype(type(el[0])))
else:
return _check_dtype(np.dtype(type(el)))
# Output types are floats, due to both softmaxes and regression req.
if n_classes is not None and n_classes > 0:
if n_classes is not None and (y is None or not y_is_dict) and n_classes > 0:
self._output_dtype = np.float32
elif y is not None:
self._output_dtype = _check_dtype(y_first_el.dtype).as_numpy_dtype
else:
self._output_dtype = dict([(k, check_y_dtype(v)) for k, v in list(y_first_el.items())]) if y_is_dict \
else (check_y_dtype(y_first_el) if y is not None else None)
def get_feed_params(self):
"""Function returns a dict with data feed params while training.
"""Function returns a `dict` with data feed params while training.
Returns:
A dict with data feed params while training.
A `dict` with data feed params while training.
"""
return {'batch_size': self._batch_size}
@ -494,50 +623,76 @@ class StreamingDataFeeder(DataFeeder):
"""Samples data and provides it to placeholders.
Returns:
Dict of input and output tensors.
`dict` of input and output tensors.
"""
def init_array(shape, dtype):
if shape is None:
return None
else:
return dict([(k, np.zeros(shape[k], dtype[k])) for k in list(shape.keys())]) if isinstance(shape, dict) else \
np.zeros(shape, dtype=dtype)
def put_data_array(dest, index, source=None, n_classes=None):
if source is None:
dest = dest[:index, :]
elif n_classes is not None and n_classes > 1:
if len(self.output_shape) == 2:
dest.itemset((index, source), 1.0)
else:
for idx, value in enumerate(source):
dest.itemset(tuple([index, idx, value]), 1.0)
else:
if len(dest.shape) > 1:
dest[index, :] = source
else:
dest[index] = source[0] if isinstance(source, list) else source
return dest
def put_data_array_or_dict(holder, index, data=None, n_classes=None):
if holder is None:
return None
if isinstance(holder, dict):
assert (isinstance(data, dict))
for k, v in list(holder.items()):
num_classes = n_classes[k] if (n_classes is not None and k in n_classes) else None
holder[k] = put_data_array(holder[k], index, data[k], num_classes)
else:
holder = put_data_array(holder, index, data, n_classes)
return holder
if self.stopped:
raise StopIteration
try:
inp = np.zeros(self.input_shape, dtype=self._input_dtype)
except TypeError as exc:
raise TypeError('Unrecognized dtype: {}. {}'.format(
self._input_dtype, exc))
if self._y is not None:
out = np.zeros(self.output_shape, dtype=self._output_dtype)
inp = init_array(self.input_shape, self._input_dtype)
out = init_array(self.output_shape, self._output_dtype)
for i in xrange(self._batch_size):
# Add handling when queue ends.
try:
inp[i, :] = six.next(self._x)
next_inp = six.next(self._x)
inp = put_data_array_or_dict(inp, i, next_inp, None)
except StopIteration:
self.stopped = True
if i == 0:
raise
inp = inp[:i, :]
if self._y is not None:
out = out[:i]
inp = put_data_array_or_dict(inp, i, None, None)
out = put_data_array_or_dict(out, i, None, None)
break
if self._y is not None:
y = six.next(self._y)
if self.n_classes is not None and self.n_classes > 1:
if len(self.output_shape) == 2:
out.itemset((i, y), 1.0)
else:
for idx, value in enumerate(y):
out.itemset(tuple([i, idx, value]), 1.0)
else:
# The y itertor can sometimes return scalars or singleton lists.
try:
out[i] = y
except ValueError as _:
assert len(y) == 1, ('Expected singleton label, got {}'
.format(repr(y)))
out[i] = y[0]
if self._y is None:
return {self._input_placeholder.name: inp}
return {self._input_placeholder.name: inp,
self._output_placeholder.name: out}
next_out = six.next(self._y)
out = put_data_array_or_dict(out, i, next_out, self.n_classes)
# creating feed_dict
feed_dict = dict([(self._input_placeholder[k].name, inp[k]) for k in list(self._input_placeholder.keys())]) if \
isinstance(inp, dict) else {self._input_placeholder.name: inp}
if self._y is not None:
feed_dict.update(
dict([(self._output_placeholder[k].name, out[k]) for k in list(self._output_placeholder.keys())]) \
if isinstance(out, dict) else {self._output_placeholder.name: out})
return feed_dict
return _feed_dict_fn
@ -575,6 +730,10 @@ class DaskDataFeeder(object):
input_dtype: dtype of input.
output_dtype: dtype of output.
"""
if isinstance(x, dict) or isinstance(y, dict):
raise ValueError("DaskDataFeeder does not support dictionaries at the moment.")
# pylint: disable=invalid-name,super-init-not-called
import dask.dataframe as dd # pylint: disable=g-import-not-at-top
# TODO(terrytangyuan): check x and y dtypes in dask_io like pandas
@ -601,7 +760,7 @@ class DaskDataFeeder(object):
self._shuffle = shuffle
self.epochs = epochs
self.input_shape, self.output_shape, self._batch_size = _get_in_out_shape(
x_shape, y_shape, n_classes, batch_size)
x_shape, y_shape, n_classes, batch_size)
self.sample_fraction = self._batch_size / float(x_count)
self._input_dtype = _check_dtype(self._x.dtypes[0])
self._output_dtype = _check_dtype(self._y.dtypes[self._y_columns])
@ -611,10 +770,10 @@ class DaskDataFeeder(object):
self.random_state = random_state
def get_feed_params(self):
"""Function returns a dict with data feed params while training.
"""Function returns a `dict` with data feed params while training.
Returns:
A dict with data feed params while training.
A `dict` with data feed params while training.
"""
return {'batch_size': self._batch_size}
@ -629,13 +788,14 @@ class DaskDataFeeder(object):
A function that when called samples a random subset of batch size
from x and y.
"""
def _feed_dict_fn():
"""Samples data and provides it to placeholders."""
# TODO(ipolosukhin): option for with/without replacement (dev version of
# dask)
sample = self.df.random_split(
[self.sample_fraction, 1 - self.sample_fraction],
random_state=self.random_state)
[self.sample_fraction, 1 - self.sample_fraction],
random_state=self.random_state)
inp = extract_pandas_matrix(sample[0][self._x_columns].compute()).tolist()
out = extract_pandas_matrix(sample[0][self._y_columns].compute())
# convert to correct dtype
@ -650,4 +810,5 @@ class DaskDataFeeder(object):
encoded_out[np.arange(out.size), out] = 1
return {input_placeholder.name: inp,
output_placeholder.name: encoded_out}
return _feed_dict_fn

250
tensorflow/contrib/learn/python/learn/learn_io/data_feeder_test.py
View File

@ -32,150 +32,200 @@ class DataFeederTest(tf.test.TestCase):
# pylint: disable=undefined-variable
"""Tests for `DataFeeder`."""
def _wrap_dict(self, data, prepend=''):
return {prepend+'1': data, prepend+'2': data}
def _assert_raises(self, input_data):
with self.assertRaisesRegexp(TypeError, 'annot convert'):
data_feeder.DataFeeder(input_data, None, n_classes=0, batch_size=1)
def test_input_uint32(self):
self._assert_raises(np.matrix([[1, 2], [3, 4]], dtype=np.uint32))
data = np.matrix([[1, 2], [3, 4]], dtype=np.uint32)
self._assert_raises(data)
self._assert_raises(self._wrap_dict(data))
def test_input_uint64(self):
self._assert_raises(np.matrix([[1, 2], [3, 4]], dtype=np.uint64))
data = np.matrix([[1, 2], [3, 4]], dtype=np.uint64)
self._assert_raises(data)
self._assert_raises(self._wrap_dict(data))
def _assert_dtype(self, expected_np_dtype, expected_tf_dtype, input_data):
feeder = data_feeder.DataFeeder(input_data, None, n_classes=0, batch_size=1)
self.assertEqual(expected_np_dtype, feeder.input_dtype)
if isinstance(input_data, dict):
for k, v in list(feeder.input_dtype.items()):
self.assertEqual(expected_np_dtype, v)
else:
self.assertEqual(expected_np_dtype, feeder.input_dtype)
with tf.Graph().as_default() as g, self.test_session(g):
inp, _ = feeder.input_builder()
self.assertEqual(expected_tf_dtype, inp.dtype)
if isinstance(inp, dict):
for k, v in list(inp.items()):
self.assertEqual(expected_tf_dtype, v.dtype)
else:
self.assertEqual(expected_tf_dtype, inp.dtype)
def test_input_int8(self):
self._assert_dtype(
np.int8, tf.int8, np.matrix([[1, 2], [3, 4]], dtype=np.int8))
data = np.matrix([[1, 2], [3, 4]], dtype=np.int8)
self._assert_dtype(np.int8, tf.int8, data)
self._assert_dtype(np.int8, tf.int8, self._wrap_dict(data))
def test_input_int16(self):
self._assert_dtype(
np.int16, tf.int16, np.matrix([[1, 2], [3, 4]], dtype=np.int16))
data = np.matrix([[1, 2], [3, 4]], dtype=np.int16)
self._assert_dtype(np.int16, tf.int16, data)
self._assert_dtype(np.int16, tf.int16, self._wrap_dict(data))
def test_input_int32(self):
self._assert_dtype(
np.int32, tf.int32, np.matrix([[1, 2], [3, 4]], dtype=np.int32))
data = np.matrix([[1, 2], [3, 4]], dtype=np.int32)
self._assert_dtype(np.int32, tf.int32, data)
self._assert_dtype(np.int32, tf.int32, self._wrap_dict(data))
def test_input_int64(self):
self._assert_dtype(
np.int64, tf.int64, np.matrix([[1, 2], [3, 4]], dtype=np.int64))
data = np.matrix([[1, 2], [3, 4]], dtype=np.int64)
self._assert_dtype(np.int64, tf.int64, data)
self._assert_dtype(np.int64, tf.int64, self._wrap_dict(data))
def test_input_uint8(self):
self._assert_dtype(
np.uint8, tf.uint8, np.matrix([[1, 2], [3, 4]], dtype=np.uint8))
data = np.matrix([[1, 2], [3, 4]], dtype=np.uint8)
self._assert_dtype(np.uint8, tf.uint8, data)
self._assert_dtype(np.uint8, tf.uint8, self._wrap_dict(data))
def test_input_uint16(self):
self._assert_dtype(
np.uint16, tf.uint16, np.matrix([[1, 2], [3, 4]], dtype=np.uint16))
data = np.matrix([[1, 2], [3, 4]], dtype=np.uint16)
self._assert_dtype(np.uint16, tf.uint16, data)
self._assert_dtype(np.uint16, tf.uint16, self._wrap_dict(data))
def test_input_float16(self):
self._assert_dtype(
np.float16, tf.float16, np.matrix([[1, 2], [3, 4]], dtype=np.float16))
data = np.matrix([[1, 2], [3, 4]], dtype=np.float16)
self._assert_dtype(np.float16, tf.float16, data)
self._assert_dtype(np.float16, tf.float16, self._wrap_dict(data))
def test_input_float32(self):
self._assert_dtype(
np.float32, tf.float32, np.matrix([[1, 2], [3, 4]], dtype=np.float32))
data = np.matrix([[1, 2], [3, 4]], dtype=np.float32)
self._assert_dtype(np.float32, tf.float32, data)
self._assert_dtype(np.float32, tf.float32, self._wrap_dict(data))
def test_input_float64(self):
self._assert_dtype(
np.float64, tf.float64, np.matrix([[1, 2], [3, 4]], dtype=np.float64))
data = np.matrix([[1, 2], [3, 4]], dtype=np.float64)
self._assert_dtype(np.float64, tf.float64, data)
self._assert_dtype(np.float64, tf.float64, self._wrap_dict(data))
def test_input_bool(self):
self._assert_dtype(
np.bool, tf.bool,
np.array([[False for _ in xrange(2)] for _ in xrange(2)]))
data = np.array([[False for _ in xrange(2)] for _ in xrange(2)])
self._assert_dtype(np.bool, tf.bool, data)
self._assert_dtype(np.bool, tf.bool, self._wrap_dict(data))
def test_input_string(self):
input_data = np.array([['str%d' % i for i in xrange(2)] for _ in xrange(2)])
self._assert_dtype(input_data.dtype, tf.string, input_data)
self._assert_dtype(input_data.dtype, tf.string, self._wrap_dict(input_data))
def _assertAllClose(self, src, dest, src_key_of=None, src_prop=None):
def func(x):
val = getattr(x, src_prop) if src_prop else x
return val if src_key_of is None else src_key_of[val]
if isinstance(src, dict):
for k in list(src.keys()):
self.assertAllClose(func(src[k]), dest)
else:
self.assertAllClose(func(src), dest)
def test_unsupervised(self):
def func(feeder):
with self.test_session():
inp, _ = feeder.input_builder()
feed_dict_fn = feeder.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self._assertAllClose(inp, [[1, 2]], feed_dict, 'name')
data = np.matrix([[1, 2], [2, 3], [3, 4]])
feeder = data_feeder.DataFeeder(data, None, n_classes=0, batch_size=1)
with self.test_session():
inp, _ = feeder.input_builder()
feed_dict_fn = feeder.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[inp.name], [[1, 2]])
func(data_feeder.DataFeeder(data, None, n_classes=0, batch_size=1))
func(data_feeder.DataFeeder(self._wrap_dict(data), None, n_classes=0, batch_size=1))
def test_data_feeder_regression(self):
def func(df):
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self._assertAllClose(inp, [[3, 4], [1, 2]], feed_dict, 'name')
self._assertAllClose(out, [2, 1], feed_dict, 'name')
x = np.matrix([[1, 2], [3, 4]])
y = np.array([1, 2])
df = data_feeder.DataFeeder(x, y, n_classes=0, batch_size=3)
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[inp.name], [[3, 4], [1, 2]])
self.assertAllClose(feed_dict[out.name], [2, 1])
func(data_feeder.DataFeeder(x, y, n_classes=0, batch_size=3))
func(data_feeder.DataFeeder(self._wrap_dict(x, 'in'), self._wrap_dict(y, 'out'),
n_classes=self._wrap_dict(0, 'out'), batch_size=3))
def test_epoch(self):
def func(feeder):
with self.test_session():
feeder.input_builder()
epoch = feeder.make_epoch_variable()
feed_dict_fn = feeder.get_feed_dict_fn()
# First input
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [0])
# Second input
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [0])
# Third input
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [0])
# Back to the first input again, so new epoch.
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [1])
data = np.matrix([[1, 2], [2, 3], [3, 4]])
labels = np.array([0, 0, 1])
feeder = data_feeder.DataFeeder(data, labels, n_classes=0, batch_size=1)
with self.test_session():
feeder.input_builder()
epoch = feeder.make_epoch_variable()
feed_dict_fn = feeder.get_feed_dict_fn()
# First input
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [0])
# Second input
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [0])
# Third input
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [0])
# Back to the first input again, so new epoch.
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[epoch.name], [1])
func(data_feeder.DataFeeder(data, labels, n_classes=0, batch_size=1))
func(data_feeder.DataFeeder(self._wrap_dict(data, 'in'), self._wrap_dict(labels, 'out'),
n_classes=self._wrap_dict(0, 'out'), batch_size=1))
def test_data_feeder_multioutput_regression(self):
def func(df):
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self._assertAllClose(inp, [[3, 4], [1, 2]], feed_dict, 'name')
self._assertAllClose(out, [[3, 4], [1, 2]], feed_dict, 'name')
x = np.matrix([[1, 2], [3, 4]])
y = np.array([[1, 2], [3, 4]])
df = data_feeder.DataFeeder(x, y, n_classes=0, batch_size=2)
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[inp.name], [[3, 4], [1, 2]])
self.assertAllClose(feed_dict[out.name], [[3, 4], [1, 2]])
func(data_feeder.DataFeeder(x, y, n_classes=0, batch_size=2))
func(data_feeder.DataFeeder(self._wrap_dict(x, 'in'), self._wrap_dict(y, 'out'),
n_classes=self._wrap_dict(0, 'out'), batch_size=2))
def test_data_feeder_multioutput_classification(self):
def func(df):
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self._assertAllClose(inp, [[3, 4], [1, 2]], feed_dict, 'name')
self._assertAllClose(out,
[[[0, 0, 1, 0, 0], [0, 0, 0, 1, 0], [0, 0, 0, 0, 1]],
[[1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, 1, 0, 0]]],
feed_dict, 'name')
x = np.matrix([[1, 2], [3, 4]])
y = np.array([[0, 1, 2], [2, 3, 4]])
df = data_feeder.DataFeeder(x, y, n_classes=5, batch_size=2)
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[inp.name], [[3, 4], [1, 2]])
self.assertAllClose(feed_dict[out.name],
[[[0, 0, 1, 0, 0], [0, 0, 0, 1, 0], [0, 0, 0, 0, 1]],
[[1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, 1, 0, 0]]])
func(data_feeder.DataFeeder(x, y, n_classes=5, batch_size=2))
func(data_feeder.DataFeeder(self._wrap_dict(x, 'in'), self._wrap_dict(y, 'out'),
n_classes=self._wrap_dict(5, 'out'), batch_size=2))
def test_streaming_data_feeder(self):
def func(df):
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self._assertAllClose(inp, [[1, 2], [3, 4]], feed_dict, 'name')
self._assertAllClose(out, [1, 2], feed_dict, 'name' )
def x_iter():
yield np.array([1, 2])
yield np.array([3, 4])
def x_iter(wrap_dict=False):
yield np.array([1, 2]) if not wrap_dict else self._wrap_dict(np.array([1, 2]), 'in')
yield np.array([3, 4]) if not wrap_dict else self._wrap_dict(np.array([3, 4]), 'in')
def y_iter():
yield np.array([1])
yield np.array([2])
def y_iter(wrap_dict=False):
yield np.array([1]) if not wrap_dict else self._wrap_dict(np.array([1]), 'out')
yield np.array([2]) if not wrap_dict else self._wrap_dict(np.array([2]), 'out')
df = data_feeder.StreamingDataFeeder(x_iter(),
y_iter(),
n_classes=0,
batch_size=2)
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[inp.name], [[1, 2], [3, 4]])
self.assertAllClose(feed_dict[out.name], [1, 2])
func(data_feeder.StreamingDataFeeder(x_iter(), y_iter(), n_classes=0, batch_size=2))
func(data_feeder.StreamingDataFeeder(x_iter(True), y_iter(True),
n_classes=self._wrap_dict(0, 'out'), batch_size=2))
def test_dask_data_feeder(self):
if HAS_PANDAS and HAS_DASK:
@ -196,6 +246,13 @@ class DataFeederTest(tf.test.TestCase):
self.assertAllClose(feed_dict[out.name], [[0., 0., 1.], [0., 1., 0.]])
def test_hdf5_data_feeder(self):
def func(df):
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self._assertAllClose(inp, [[3, 4], [1, 2]], feed_dict, 'name')
self.assertAllClose(out, [2, 1], feed_dict, 'name')
try:
import h5py # pylint: disable=g-import-not-at-top
x = np.matrix([[1, 2], [3, 4]])
@ -207,25 +264,28 @@ class DataFeederTest(tf.test.TestCase):
h5f = h5py.File('test_hdf5.h5', 'r')
x = h5f['x']
y = h5f['y']
df = data_feeder.DataFeeder(x, y, n_classes=0, batch_size=3)
inp, out = df.input_builder()
feed_dict_fn = df.get_feed_dict_fn()
feed_dict = feed_dict_fn()
self.assertAllClose(feed_dict[inp.name], [[3, 4], [1, 2]])
self.assertAllClose(feed_dict[out.name], [2, 1])
func(data_feeder.DataFeeder(x, y, n_classes=0, batch_size=3))
func(data_feeder.DataFeeder(self._wrap_dict(x, 'in'), self._wrap_dict(y, 'out'),
n_classes=self._wrap_dict(0, 'out'), batch_size=3))
except ImportError:
print("Skipped test for hdf5 since it's not installed.")
class SetupPredictDataFeederTest(tf.test.TestCase):
class SetupPredictDataFeederTest(DataFeederTest):
"""Tests for `DataFeeder.setup_predict_data_feeder`."""
def test_iterable_data(self):
# pylint: disable=undefined-variable
x = iter([[1, 2], [3, 4], [5, 6]])
df = data_feeder.setup_predict_data_feeder(x, batch_size=2)
self.assertAllClose(six.next(df), [[1, 2], [3, 4]])
self.assertAllClose(six.next(df), [[5, 6]])
def func(df):
self._assertAllClose(six.next(df), [[1, 2], [3, 4]])
self._assertAllClose(six.next(df), [[5, 6]])
data = [[1, 2], [3, 4], [5, 6]]
x = iter(data)
x_dict = iter([self._wrap_dict(v) for v in iter(data)])
func(data_feeder.setup_predict_data_feeder(x, batch_size=2))
func(data_feeder.setup_predict_data_feeder(x_dict, batch_size=2))
if __name__ == '__main__':

17
tensorflow/contrib/learn/python/learn/trainable.py
View File

@ -33,17 +33,17 @@ class Trainable(object):
"""Trains a model given training data `x` predictions and `y` labels.
Args:
x: Matrix of shape [n_samples, n_features...]. Can be iterator that
returns arrays of features. The training input samples for fitting the
model. If set, `input_fn` must be `None`.
y: Vector or matrix [n_samples] or [n_samples, n_outputs]. Can be
iterator that returns array of labels. The training label values
(class labels in classification, real numbers in regression). If set,
`input_fn` must be `None`. Note: For classification, label values must
x: Matrix of shape [n_samples, n_features...] or the dictionary of Matrices.
Can be iterator that returns arrays of features or dictionary of arrays of features.
The training input samples for fitting the model. If set, `input_fn` must be `None`.
y: Vector or matrix [n_samples] or [n_samples, n_outputs] or the dictionary of same.
Can be iterator that returns array of labels or dictionary of array of labels.
The training label values (class labels in classification, real numbers in regression).
If set, `input_fn` must be `None`. Note: For classification, label values must
be integers representing the class index (i.e. values from 0 to
n_classes-1).
input_fn: Input function returning a tuple of:
features - Dictionary of string feature name to `Tensor` or `Tensor`.
features - `Tensor` or dictionary of string feature name to `Tensor`.
labels - `Tensor` or dictionary of `Tensor` with labels.
If input_fn is set, `x`, `y`, and `batch_size` must be `None`.
steps: Number of steps for which to train model. If `None`, train forever.
@ -67,4 +67,3 @@ class Trainable(object):
`self`, for chaining.
"""
raise NotImplementedError

1
tensorflow/contrib/learn/python/learn/utils/__init__.py
View File

@ -19,5 +19,4 @@ from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.contrib.learn.python.learn.utils import checkpoints
from tensorflow.contrib.learn.python.learn.utils.export import export_estimator

3
tensorflow/contrib/makefile/README.md
View File

@ -142,10 +142,11 @@ xcode-select --install
If this is a new install, you will need to run XCode once to agree to the
license before continuing.
Then install [automake](https://en.wikipedia.org/wiki/Automake):
Then install [automake](https://en.wikipedia.org/wiki/Automake)/[libtool](https://en.wikipedia.org/wiki/GNU_Libtool):
```bash
brew install automake
brew install libtool
```
Also, download the graph if you haven't already:

5
tensorflow/contrib/makefile/compile_ios_protobuf.sh
View File

@ -67,7 +67,6 @@ fi
make distclean
./configure \
--build=x86_64-apple-${OSX_VERSION} \
--host=i386-apple-${OSX_VERSION} \
--disable-shared \
--enable-cross-compile \
@ -95,7 +94,6 @@ make install
make distclean
./configure \
--build=x86_64-apple-${OSX_VERSION} \
--host=x86_64-apple-${OSX_VERSION} \
--disable-shared \
--enable-cross-compile \
@ -123,7 +121,6 @@ make install
make distclean
./configure \
--build=x86_64-apple-${OSX_VERSION} \
--host=armv7-apple-${OSX_VERSION} \
--with-protoc="${PROTOC_PATH}" \
--disable-shared \
@ -147,7 +144,6 @@ make install
make distclean
./configure \
--build=x86_64-apple-${OSX_VERSION} \
--host=armv7s-apple-${OSX_VERSION} \
--with-protoc="${PROTOC_PATH}" \
--disable-shared \
@ -171,7 +167,6 @@ make install
make distclean
./configure \
--build=x86_64-apple-${OSX_VERSION} \
--host=arm \
--with-protoc="${PROTOC_PATH}" \
--disable-shared \

1
tensorflow/contrib/metrics/python/ops/set_ops.py
View File

@ -19,7 +19,6 @@ from __future__ import print_function
from tensorflow.python.ops import sets
set_size = sets.set_size
set_intersection = sets.set_intersection

6
tensorflow/contrib/rnn/python/kernel_tests/gru_ops_test.py
View File

@ -158,7 +158,7 @@ class GRUBlockCellTest(tf.test.TestCase):
output = gru_ops.GRUBlockCell(cell_size)(x, h)
sess.run([tf.global_variables_initializer()])
all_variables = tf.all_variables()[0:4]
all_variables = tf.global_variables()[0:4]
[w_ru, b_ru, w_c, b_c] = all_variables
d_new_h_wrt_x = tf.gradients([output], x)
@ -178,7 +178,7 @@ class GRUBlockCellTest(tf.test.TestCase):
output = tf.contrib.rnn.GRUCell(cell_size)(x, h)
sess.run([tf.global_variables_initializer()])
all_variables = tf.all_variables()[4:8]
all_variables = tf.global_variables()[4:8]
[w_ru, b_ru, w_c, b_c] = all_variables
d_new_h_wrt_x = tf.gradients([output], x)
@ -281,7 +281,7 @@ class GRUBlockCellTest(tf.test.TestCase):
sess.run([tf.global_variables_initializer()])
all_variables = tf.all_variables()
all_variables = tf.global_variables()
[w_ru, b_ru, w_c, b_c] = all_variables[:4]

2
tensorflow/contrib/rnn/python/kernel_tests/rnn_test.py
View File

@ -382,7 +382,7 @@ class StackBidirectionalRNNTest(tf.test.TestCase):
# check that all the variables names starts with the proper scope.
tf.global_variables_initializer()
all_vars = tf.all_variables()
all_vars = tf.global_variables()
prefix = prefix or "stack_bidirectional_rnn"
scope_vars = [v for v in all_vars if v.name.startswith(prefix + "/")]
tf.logging.info("StackRNN with scope: %s (%s)"

8
tensorflow/contrib/seq2seq/BUILD
View File

@ -16,9 +16,9 @@ py_library(
)
cuda_py_test(
name = "layers_test",
name = "decoder_fn_test",
size = "medium",
srcs = ["python/kernel_tests/layers_test.py"],
srcs = ["python/kernel_tests/decoder_fn_test.py"],
additional_deps = [
":seq2seq_py",
"//tensorflow:tensorflow_py",
@ -28,9 +28,9 @@ cuda_py_test(
)
cuda_py_test(
name = "loss_test",
name = "seq2seq_test",
size = "medium",
srcs = ["python/kernel_tests/loss_test.py"],
srcs = ["python/kernel_tests/seq2seq_test.py"],
additional_deps = [
":seq2seq_py",
"//tensorflow:tensorflow_py",

7
tensorflow/contrib/seq2seq/__init__.py
View File

@ -13,7 +13,7 @@
# limitations under the License.
# ==============================================================================
"""Ops for building neural network seq2seq layers and losses."""
"""Ops for building neural network seq2seq decoders and losses."""
from __future__ import absolute_import
from __future__ import division
@ -22,6 +22,7 @@ from __future__ import print_function
import sys
# pylint: disable=unused-import,line-too-long
from tensorflow.contrib.seq2seq.python.ops import layers
from tensorflow.contrib.seq2seq.python.ops import loss
from tensorflow.contrib.seq2seq.python.ops.decoder_fn import *
from tensorflow.contrib.seq2seq.python.ops.loss import *
from tensorflow.contrib.seq2seq.python.ops.seq2seq import *
# pylint: enable=unused-import,line-too-long

0
tensorflow/contrib/seq2seq/python/kernel_tests/__init__.py Normal file
View File

9
tensorflow/contrib/seq2seq/python/kernel_tests/layers_test.py → tensorflow/contrib/seq2seq/python/kernel_tests/decoder_fn_test.py
View File

@ -13,7 +13,7 @@
# limitations under the License.
# ==============================================================================
"""Tests for contrib.seq2seq.python.seq2seq.layers_ops."""
"""Tests for contrib.seq2seq.python.seq2seq.loss_ops."""
# pylint: disable=unused-import,g-bad-import-order
from __future__ import absolute_import
from __future__ import division
@ -23,12 +23,9 @@ from __future__ import print_function
import tensorflow as tf
class LayersTest(tf.test.TestCase):
class DecoderFnTest(tf.test.TestCase):
def testRNNDecoder(self):
pass
def testRNNDecoderAttention(self):
def testDecoderFn(self):
pass

129
tensorflow/contrib/seq2seq/python/kernel_tests/seq2seq_test.py Normal file
View File

@ -0,0 +1,129 @@
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Tests for contrib.seq2seq.python.ops.seq2seq."""
# pylint: disable=unused-import,g-bad-import-order
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
# pylint: enable=unused-import
import tensorflow as tf
from tensorflow.contrib import layers
class Seq2SeqTest(tf.test.TestCase):
# test a default call of rnn_decoder
def test_rnn_decoder(self):
pass
# test default call with time_major=True
def test_dynamic_rnn_decoder_time_major(self):
with self.test_session() as sess:
with tf.variable_scope("root", initializer=
tf.constant_initializer(0.5)) as varscope:
# Define inputs/outputs to model
batch_size = 2
encoder_embedding_size = 3
decoder_embedding_size = 4
encoder_hidden_size = 5
decoder_hidden_size = encoder_hidden_size
input_sequence_length = 6
decoder_sequence_length = 7
num_decoder_symbols = 20
start_of_sequence_id = end_of_sequence_id = 1
decoder_embeddings = tf.get_variable('decoder_embeddings',
[num_decoder_symbols, decoder_embedding_size],
initializer=tf.random_normal_initializer(stddev=0.1))
inputs = tf.constant(0.5, shape=[input_sequence_length, batch_size,
encoder_embedding_size])
decoder_inputs = tf.constant(0.4, shape=[decoder_sequence_length,
batch_size,
decoder_embedding_size])
decoder_length = tf.constant(decoder_sequence_length, dtype=tf.int32,
shape=[batch_size,])
with tf.variable_scope("rnn") as scope:
# setting up weights for computing the final output
output_fn = lambda x: layers.linear(x, num_decoder_symbols,
scope=scope)
# Define model
encoder_outputs, encoder_state = tf.nn.dynamic_rnn(
cell=tf.nn.rnn_cell.GRUCell(encoder_hidden_size), inputs=inputs,
dtype=tf.float32, time_major=True, scope=scope)
with tf.variable_scope("decoder") as scope:
# Train decoder
decoder_cell = tf.nn.rnn_cell.GRUCell(decoder_hidden_size)
decoder_fn_train = tf.contrib.seq2seq.simple_decoder_fn_train(
encoder_state=encoder_state)
decoder_outputs_train, decoder_state_train = (
tf.contrib.seq2seq.dynamic_rnn_decoder(
cell=decoder_cell,
decoder_fn=decoder_fn_train,
inputs=decoder_inputs,
sequence_length=decoder_length,
time_major=True,
scope=scope))
decoder_outputs_train = output_fn(decoder_outputs_train)
# Setup variable reuse
scope.reuse_variables()
# Inference decoder
decoder_fn_inference = (
tf.contrib.seq2seq.simple_decoder_fn_inference(
output_fn=output_fn,
encoder_state=encoder_state,
embeddings=decoder_embeddings,
start_of_sequence_id=start_of_sequence_id,
end_of_sequence_id=end_of_sequence_id,
#TODO: find out why it goes to +1
maximum_length=decoder_sequence_length-1,
num_decoder_symbols=num_decoder_symbols,
dtype=tf.int32))
decoder_outputs_inference, decoder_state_inference = (
tf.contrib.seq2seq.dynamic_rnn_decoder(
cell=decoder_cell,
decoder_fn=decoder_fn_inference,
time_major=True,
scope=scope))
# Run model
tf.global_variables_initializer().run()
decoder_outputs_train_res, decoder_state_train_res = sess.run(
[decoder_outputs_train, decoder_state_train])
decoder_outputs_inference_res, decoder_state_inference_res = sess.run(
[decoder_outputs_inference, decoder_state_inference])
# Assert outputs
self.assertEqual((decoder_sequence_length, batch_size,
num_decoder_symbols),
decoder_outputs_train_res.shape)
self.assertEqual((batch_size, num_decoder_symbols),
decoder_outputs_inference_res.shape[1:3])
self.assertEqual((batch_size, decoder_hidden_size),
decoder_state_train_res.shape)
self.assertEqual((batch_size, decoder_hidden_size),
decoder_state_inference_res.shape)
# The dynamic decoder might end earlier than `maximal_length`
# under inference
true_value = (decoder_sequence_length>=
decoder_state_inference_res.shape[0])
self.assertEqual((true_value), True)
if __name__ == '__main__':
tf.test.main()

0
tensorflow/contrib/seq2seq/python/ops/__init__.py Normal file
View File

249
tensorflow/contrib/seq2seq/python/ops/decoder_fn.py Normal file
View File

@ -0,0 +1,249 @@
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Seq2seq loss operations for use in neural networks.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.python.framework import dtypes
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import rnn_cell
from tensorflow.python.util import nest
__all__ = ["simple_decoder_fn_train",
"simple_decoder_fn_inference"]
def simple_decoder_fn_train(encoder_state, name=None):
""" Simple decoder function for a sequence-to-sequence model used in the
`dynamic_rnn_decoder`.
The `simple_decoder_fn_train` is a simple training function for a
sequence-to-sequence model. It should be used when `dynamic_rnn_decoder` is
in the training mode.
The `simple_decoder_fn_train` is called with a set of the user arguments and
returns the `decoder_fn`, which can be passed to the `dynamic_rnn_decoder`,
such that
```
dynamic_fn_train = simple_decoder_fn_train(encoder_state)
outputs_train, state_train = dynamic_rnn_decoder(
decoder_fn=dynamic_fn_train, ...)
```
Further usage can be found in the `kernel_tests/seq2seq_test.py`.
Args:
encoder_state: The encoded state to initialize the `dynamic_rnn_decoder`.
name: (default: `None`) NameScope for the decoder function;
defaults to "simple_decoder_fn_train"
Returns:
A decoder function with the required interface of `dynamic_rnn_decoder`
intended for training.
"""
with ops.name_scope(name, "simple_decoder_fn_train", [encoder_state]):
pass
def decoder_fn(time, cell_state, cell_input, cell_output, context_state):
""" Decoder function used in the `dynamic_rnn_decoder` with the purpose of
training.
Args:
time: positive integer constant reflecting the current timestep.
cell_state: state of RNNCell.
cell_input: input provided by `dynamic_rnn_decoder`.
cell_output: output of RNNCell.
context_state: context state provided by `dynamic_rnn_decoder`.
Returns:
A tuple (done, next state, next input, emit output, next context state)
where:
done: `None`, which is used by the `dynamic_rnn_decoder` to indicate
that `sequence_lengths` in `dynamic_rnn_decoder` should be used.
next state: `cell_state`, this decoder function does not modify the
given state.
next input: `cell_input`, this decoder function does not modify the
given input. The input could be modified when applying e.g. attention.
emit output: `cell_output`, this decoder function does not modify the
given output.
next context state: `context_state`, this decoder function does not
modify the given context state. The context state could be modified when
applying e.g. beam search.
"""
with ops.name_scope(name, "simple_decoder_fn_train",
[time, cell_state, cell_input, cell_output,
context_state]):
if cell_state is None: # first call, return encoder_state
return (None, encoder_state, cell_input, cell_output, context_state)
else:
return (None, cell_state, cell_input, cell_output, context_state)
return decoder_fn
def simple_decoder_fn_inference(output_fn, encoder_state, embeddings,
start_of_sequence_id, end_of_sequence_id,
maximum_length, num_decoder_symbols,
dtype=dtypes.int32, name=None):
""" Simple decoder function for a sequence-to-sequence model used in the
`dynamic_rnn_decoder`.
The `simple_decoder_fn_inference` is a simple inference function for a
sequence-to-sequence model. It should be used when `dynamic_rnn_decoder` is
in the inference mode.
The `simple_decoder_fn_inference` is called with a set of the user arguments
and returns the `decoder_fn`, which can be passed to the
`dynamic_rnn_decoder`, such that
```
dynamic_fn_inference = simple_decoder_fn_inference(...)
outputs_inference, state_inference = dynamic_rnn_decoder(
decoder_fn=dynamic_fn_inference, ...)
```
Further usage can be found in the `kernel_tests/seq2seq_test.py`.
Args:
output_fn: An output function to project your `cell_output` onto class
logits.
An example of an output function;
```
tf.variable_scope("decoder") as varscope
output_fn = lambda x: layers.linear(x, num_decoder_symbols,
scope=varscope)
outputs_train, state_train = seq2seq.dynamic_rnn_decoder(...)
logits_train = output_fn(outputs_train)
varscope.reuse_variables()
logits_inference, state_inference = seq2seq.dynamic_rnn_decoder(
output_fn=output_fn, ...)
```
If `None` is supplied it will act as an identity function, which
might be wanted when using the RNNCell `OutputProjectionWrapper`.
encoder_state: The encoded state to initialize the `dynamic_rnn_decoder`.
embeddings: The embeddings matrix used for the decoder sized
`[num_decoder_symbols, embedding_size]`.
start_of_sequence_id: The start of sequence ID in the decoder embeddings.
end_of_sequence_id: The end of sequence ID in the decoder embeddings.
maximum_length: The maximum allowed of time steps to decode.
num_decoder_symbols: The number of classes to decode at each time step.
dtype: (default: `dtypes.int32`) The default data type to use when
handling integer objects.
name: (default: `None`) NameScope for the decoder function;
defaults to "simple_decoder_fn_inference"
Returns:
A decoder function with the required interface of `dynamic_rnn_decoder`
intended for inference.
"""
with ops.name_scope(name, "simple_decoder_fn_inference",
[output_fn, encoder_state, embeddings,
start_of_sequence_id, end_of_sequence_id,
maximum_length, num_decoder_symbols, dtype]):
start_of_sequence_id = ops.convert_to_tensor(start_of_sequence_id, dtype)
end_of_sequence_id = ops.convert_to_tensor(end_of_sequence_id, dtype)
maximum_length = ops.convert_to_tensor(maximum_length, dtype)
num_decoder_symbols = ops.convert_to_tensor(num_decoder_symbols, dtype)
encoder_info = nest.flatten(encoder_state)[0]
batch_size = encoder_info.get_shape()[0].value
if output_fn is None:
output_fn = lambda x: x
if batch_size is None:
batch_size = array_ops.shape(encoder_info)[0]
def decoder_fn(time, cell_state, cell_input, cell_output, context_state):
""" Decoder function used in the `dynamic_rnn_decoder` with the purpose of
inference.
The main difference between this decoder function and the `decoder_fn` in
`simple_decoder_fn_train` is how `next_cell_input` is calculated. In this
decoder function we calculate the next input by applying an argmax across
the feature dimension of the output from the decoder. This is a
greedy-search approach. (Bahdanau et al., 2014) & (Sutskever et al., 2014)
use beam-search instead.
Args:
time: positive integer constant reflecting the current timestep.
cell_state: state of RNNCell.
cell_input: input provided by `dynamic_rnn_decoder`.
cell_output: output of RNNCell.
context_state: context state provided by `dynamic_rnn_decoder`.
Returns:
A tuple (done, next state, next input, emit output, next context state)
where:
done: A boolean vector to indicate which sentences has reached a
`end_of_sequence_id`. This is used for early stopping by the
`dynamic_rnn_decoder`. When `time>=maximum_length` a boolean vector with
all elements as `true` is returned.
next state: `cell_state`, this decoder function does not modify the
given state.
next input: The embedding from argmax of the `cell_output` is used as
`next_input`.
emit output: If `output_fn is None` the supplied `cell_output` is
returned, else the `output_fn` is used to update the `cell_output`
before calculating `next_input` and returning `cell_output`.
next context state: `context_state`, this decoder function does not
modify the given context state. The context state could be modified when
applying e.g. beam search.
"""
with ops.name_scope(name, "simple_decoder_fn_inference",
[time, cell_state, cell_input, cell_output,
context_state]):
if cell_input is not None:
raise ValueError("Expected cell_input to be None, but saw: %s" %
cell_input)
if cell_output is None:
# invariant that this is time == 0
next_input_id = array_ops.ones([batch_size,], dtype=dtype) * (
start_of_sequence_id)
done = array_ops.zeros([batch_size,], dtype=dtypes.bool)
cell_state = encoder_state
cell_output = array_ops.zeros([num_decoder_symbols],
dtype=dtypes.float32)
else:
cell_output = output_fn(cell_output)
next_input_id = math_ops.cast(
math_ops.argmax(cell_output, 1), dtype=dtype)
done = math_ops.equal(next_input_id, end_of_sequence_id)
next_input = array_ops.gather(embeddings, next_input_id)
# if time > maxlen, return all true vector
done = control_flow_ops.cond(math_ops.greater(time, maximum_length),
lambda: array_ops.ones([batch_size,], dtype=dtypes.bool),
lambda: done)
return (done, cell_state, next_input, cell_output, context_state)
return decoder_fn

208
tensorflow/contrib/seq2seq/python/ops/seq2seq.py Normal file
View File

@ -0,0 +1,208 @@
# Copyright 2016 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Seq2seq layer operations for use in neural networks.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from tensorflow.contrib import layers
from tensorflow.python.framework import ops
from tensorflow.python.ops import array_ops
from tensorflow.python.ops import control_flow_ops
from tensorflow.python.ops import math_ops
from tensorflow.python.ops import rnn
from tensorflow.python.ops import tensor_array_ops
from tensorflow.python.ops import variable_scope as vs
__all__ = ["dynamic_rnn_decoder"]
def dynamic_rnn_decoder(cell, decoder_fn, inputs=None, sequence_length=None,
parallel_iterations=None, swap_memory=False,
time_major=False, scope=None, name=None):
""" Dynamic RNN decoder for a sequence-to-sequence model specified by
RNNCell and decoder function.
The `dynamic_rnn_decoder` is similar to the `tf.python.ops.rnn.dynamic_rnn`
as the decoder does not make any assumptions of sequence length and batch
size of the input.
The `dynamic_rnn_decoder` has two modes: training or inference and expects
the user to create seperate functions for each.
Under both training and inference `cell` and `decoder_fn` is expected. Where
the `cell` performs computation at every timestep using the `raw_rnn` and
the `decoder_fn` allows modelling of early stopping, output, state, and next
input and context.
When training the user is expected to supply `inputs`. At every time step a
slice of the supplied input is fed to the `decoder_fn`, which modifies and
returns the input for the next time step.
`sequence_length` is needed at training time, i.e., when `inputs` is not
None, for dynamic unrolling. At test time, when `inputs` is None,
`sequence_length` is not needed.
Under inference `inputs` is expected to be `None` and the input is inferred
solely from the `decoder_fn`.
Args:
cell: An instance of RNNCell.
decoder_fn: A function that takes time, cell state, cell input,
cell output and context state. It returns a early stopping vector,
cell state, next input, cell output and context state.
Examples of decoder_fn can be found in the decoder_fn.py folder.
inputs: The inputs for decoding (embedded format).
If `time_major == False` (default), this must be a `Tensor` of shape:
`[batch_size, max_time, ...]`.
If `time_major == True`, this must be a `Tensor` of shape:
`[max_time, batch_size, ...]`.
The input to `cell` at each time step will be a `Tensor` with dimensions
`[batch_size, ...]`.
sequence_length: (optional) An int32/int64 vector sized `[batch_size]`.
if `inputs` is not None and `sequence_length` is None it is inferred
from the `inputs` as the maximal possible sequence length.
parallel_iterations: (Default: 32). The number of iterations to run in
parallel. Those operations which do not have any temporal dependency
and can be run in parallel, will be. This parameter trades off
time for space. Values >> 1 use more memory but take less time,
while smaller values use less memory but computations take longer.
swap_memory: Transparently swap the tensors produced in forward inference
but needed for back prop from GPU to CPU. This allows training RNNs
which would typically not fit on a single GPU, with very minimal (or no)
performance penalty.
time_major: The shape format of the `inputs` and `outputs` Tensors.
If true, these `Tensors` must be shaped `[max_time, batch_size, depth]`.
If false, these `Tensors` must be shaped `[batch_size, max_time, depth]`.
Using `time_major = True` is a bit more efficient because it avoids
transposes at the beginning and end of the RNN calculation. However,
most TensorFlow data is batch-major, so by default this function
accepts input and emits output in batch-major form.
scope: VariableScope for the `raw_rnn`;
defaults to None.
name: NameScope for the decoder;
defaults to "dynamic_rnn_decoder"
Returns:
A pair (outputs, state) where:
outputs: the RNN output 'Tensor'.
If time_major == False (default), this will be a `Tensor` shaped:
`[batch_size, max_time, cell.output_size]`.
If time_major == True, this will be a `Tensor` shaped:
`[max_time, batch_size, cell.output_size]`.
state: The final state and will be shaped
`[batch_size, cell.state_size]`.
Raises:
ValueError: if inputs is not None and has less than three dimensions.
"""
with ops.name_scope(name, "dynamic_rnn_decoder",
[cell, decoder_fn, inputs, sequence_length,
parallel_iterations, swap_memory, time_major, scope]):
if inputs is not None:
# Convert to tensor
inputs = ops.convert_to_tensor(inputs)
# Test input dimensions
if inputs.get_shape().ndims is not None and (
inputs.get_shape().ndims < 2):
raise ValueError("Inputs must have at least two dimensions")
# Setup of RNN (dimensions, sizes, length, initial state, dtype)
if not time_major:
# [batch, seq, features] -> [seq, batch, features]
inputs = array_ops.transpose(inputs, perm=[1, 0, 2])
dtype = inputs.dtype
# Get data input information
input_depth = int(inputs.get_shape()[2])
batch_depth = inputs.get_shape()[1].value
max_time = inputs.get_shape()[0].value
if max_time is None:
max_time = array_ops.shape(inputs)[0]
# Setup decoder inputs as TensorArray
inputs_ta = tensor_array_ops.TensorArray(dtype, size=max_time)
inputs_ta = inputs_ta.unpack(inputs)
def loop_fn(time, cell_output, cell_state, loop_state):
if cell_state is None: # first call, before while loop (in raw_rnn)
if cell_output is not None:
raise ValueError("Expected cell_output to be None when cell_state "
"is None, but saw: %s" % cell_output)
if loop_state is not None:
raise ValueError("Expected loop_state to be None when cell_state "
"is None, but saw: %s" % loop_state)
context_state = None
else: # subsequent calls, inside while loop, after cell excution
if isinstance(loop_state, tuple):
(done, context_state) = loop_state
else:
done = loop_state
context_state = None
# call decoder function
if inputs is not None: # training
# get next_cell_input
if cell_state is None:
next_cell_input = inputs_ta.read(0)
else:
if batch_depth is not None:
batch_size = batch_depth
else:
batch_size = array_ops.shape(done)[0]
next_cell_input = control_flow_ops.cond(
math_ops.equal(time, max_time),
lambda: array_ops.zeros([batch_size, input_depth], dtype=dtype),
lambda: inputs_ta.read(time))
(next_done, next_cell_state, next_cell_input, emit_output,
next_context_state) = decoder_fn(time, cell_state, next_cell_input,
cell_output, context_state)
else: # inference
# next_cell_input is obtained through decoder_fn
(next_done, next_cell_state, next_cell_input, emit_output,
next_context_state) = decoder_fn(time, cell_state, None, cell_output,
context_state)
# check if we are done
if next_done is None: # training
next_done = time >= sequence_length
# build next_loop_state
if next_context_state is None:
next_loop_state = next_done
else:
next_loop_state = (next_done, next_context_state)
return (next_done, next_cell_input, next_cell_state,
emit_output, next_loop_state)
# Run raw_rnn function
outputs_ta, state, _ = rnn.raw_rnn(
cell, loop_fn, parallel_iterations=parallel_iterations,
swap_memory=swap_memory, scope=scope)
outputs = outputs_ta.pack()
if not time_major:
# [seq, batch, features] -> [batch, seq, features]
outputs = array_ops.transpose(outputs, perm=[1, 0, 2])
return outputs, state

2
tensorflow/contrib/session_bundle/session_bundle.cc
View File

@ -48,7 +48,7 @@ auto* load_attempt_count = monitoring::Counter<2>::New(
"model_path", "status");
auto* load_latency = monitoring::Counter<1>::New(
"/tensorflow/contrib/session_bundle/load_latency",
"Latency in microseconds for SessionBundles that were succesfully loaded.",
"Latency in microseconds for SessionBundles that were successfully loaded.",
"model_path");
constexpr char kLoadAttemptFail[] = "fail";
constexpr char kLoadAttemptSuccess[] = "success";

6
tensorflow/contrib/slim/README.md
View File

@ -109,7 +109,7 @@ weights = variables.variable('weights',
Note that in native TensorFlow, there are two types of variables: regular
variables and local (transient) variables. The vast majority of variables are
regular variables: once created, they can be saved to disk using a
[saver](https://www.tensorflow.org/versions/r0.9/api_docs/python/state_ops.html#Saver).
[saver](https://www.tensorflow.org/versions/r0.11/api_docs/python/state_ops.html#Saver).
Local variables are those variables that only exist for the duration of a
session and are not saved to disk.
@ -215,7 +215,7 @@ Dropout| [slim.dropout](https://www.tensorflow.org/code/tensorflow/contrib/layer
Flatten | [slim.flatten](https://www.tensorflow.org/code/tensorflow/contrib/layers/python/layers/layers.py)
MaxPool2D | [slim.max_pool2d](https://www.tensorflow.org/code/tensorflow/contrib/layers/python/layers/layers.py)
OneHotEncoding | [slim.one_hot_encoding](https://www.tensorflow.org/code/tensorflow/contrib/layers/python/layers/layers.py)
SeperableConv2 | [slim.seperable_conv2d](https://www.tensorflow.org/code/tensorflow/contrib/layers/python/layers/layers.py)
SeparableConv2 | [slim.separable_conv2d](https://www.tensorflow.org/code/tensorflow/contrib/layers/python/layers/layers.py)
UnitNorm | [slim.unit_norm](https://www.tensorflow.org/code/tensorflow/contrib/layers/python/layers/layers.py)
TF-Slim also provides two meta-operations called `repeat` and `stack` that
@ -901,7 +901,7 @@ slim.evaluation.evaluation_loop(
log_dir,
num_evals=num_batches,
eval_op=names_to_updates.values(),
summary_op=tf.merge_summary(summary_ops),
summary_op=tf.summary.merge(summary_ops),
eval_interval_secs=eval_interval_secs)
```

2
tensorflow/contrib/slim/python/slim/evaluation.py
View File

@ -283,5 +283,3 @@ def evaluation_loop(master,
config=session_config,
max_number_of_evaluations=max_number_of_evaluations,
timeout=timeout)

4
tensorflow/contrib/slim/python/slim/learning_test.py
View File

@ -625,7 +625,7 @@ class TrainTest(tf.test.TestCase):
tf.set_random_seed(2)
train_op = self.create_train_op()
model_variables = tf.all_variables()
model_variables = tf.global_variables()
model_path = os.path.join(logdir1, 'model.ckpt-300')
init_op = tf.global_variables_initializer()
@ -674,7 +674,7 @@ class TrainTest(tf.test.TestCase):
tf.set_random_seed(2)
train_op = self.create_train_op()
model_variables = tf.all_variables()
model_variables = tf.global_variables()
model_path = os.path.join(logdir1, 'model.ckpt-300')
saver = tf.train.Saver(model_variables)
def RestoreFn(sess):

2
tensorflow/contrib/slim/python/slim/model_analyzer.py
View File

@ -84,7 +84,7 @@ def analyze_vars(variables, print_info=False):
"""Prints the names and shapes of the variables.
Args:
variables: list of variables, for example tf.all_variables().
variables: list of variables, for example tf.global_variables().
print_info: Optional, if true print variables and their shape.
Returns:

4
tensorflow/contrib/specs/python/specs_test.py
View File

@ -197,7 +197,7 @@ class SpecsTest(tf.test.TestCase):
initializer=tf.constant_initializer(42.0))
inputs = tf.constant(_rand(10, 100))
outputs = v.funcall(inputs)
self.assertEqual(len(tf.all_variables()), 1)
self.assertEqual(len(tf.global_variables()), 1)
sess.run([outputs.initializer])
outputs_value = outputs.eval()
self.assertEqual(outputs_value.shape, (2, 2))
@ -211,7 +211,7 @@ class SpecsTest(tf.test.TestCase):
g = f | f | f | f
inputs = tf.constant(_rand(10, 100))
_ = g.funcall(inputs)
self.assertEqual(len(tf.all_variables()), 2)
self.assertEqual(len(tf.global_variables()), 2)
def testAutoFunction(self):
with self.test_session():

2
tensorflow/contrib/stat_summarizer/python/stat_summarizer_test.py
View File

@ -34,7 +34,7 @@ class StatSummarizerTest(tf.test.TestCase):
graph_def.SerializeToString())
with self.test_session() as sess:
sess.run(tf.initialize_all_variables())
sess.run(tf.global_variables_initializer())
for _ in range(20):
run_metadata = tf.RunMetadata()

6
tensorflow/contrib/tensorboard/plugins/projector/projector_api_test.py
View File

@ -31,7 +31,7 @@ class ProjectorApiTest(tf.test.TestCase):
# Create a dummy configuration.
config = tf.contrib.tensorboard.plugins.projector.ProjectorConfig()
config.model_checkpoint_path = 'test'
emb1 = config.embedding.add()
emb1 = config.embeddings.add()
emb1.tensor_name = 'tensor1'
emb1.metadata_path = 'metadata1'
@ -47,3 +47,7 @@ class ProjectorApiTest(tf.test.TestCase):
config2 = tf.contrib.tensorboard.plugins.projector.ProjectorConfig()
text_format.Parse(f.read(), config2)
self.assertEqual(config, config2)
if __name__ == "__main__":
tf.test.main()

2
tensorflow/contrib/training/python/training/device_setter.py
View File

@ -42,7 +42,7 @@ class GreedyLoadBalancingStrategy(object):
off CPU-intensive ops with RAM-intensive ops with network bandwidth.
This class is intended to be used as a `ps_strategy` in
`tf.replica_device_setter`.
`tf.train.replica_device_setter`.
"""
def __init__(self, num_tasks, load_fn):

4
tensorflow/contrib/training/python/training/evaluation_test.py
View File

@ -51,7 +51,7 @@ class CheckpointIteratorTest(tf.test.TestCase):
saver = tf.train.Saver() # Saves the global step.
with self.test_session() as session:
session.run(tf.initialize_all_variables())
session.run(tf.global_variables_initializer())
save_path = os.path.join(checkpoint_dir, 'model.ckpt')
saver.save(session, save_path, global_step=global_step)
@ -81,7 +81,7 @@ class CheckpointIteratorTest(tf.test.TestCase):
target='',
config=tf.ConfigProto(device_count={'CPU': 2})) as session:
session.run(tf.initialize_all_variables())
session.run(tf.global_variables_initializer())
save_path = os.path.join(checkpoint_dir, 'model.ckpt')
saver.save(session, save_path, global_step=global_step)

2
tensorflow/contrib/training/python/training/training_test.py
View File

@ -310,7 +310,7 @@ class TrainTest(tf.test.TestCase):
tf.set_random_seed(2)
train_op = self.create_train_op()
model_variables = tf.all_variables()
model_variables = tf.global_variables()
model_path = os.path.join(logdir1, 'model.ckpt-300')
assign_fn = tf.contrib.framework.assign_from_checkpoint_fn(

32
tensorflow/core/BUILD
View File

@ -949,12 +949,12 @@ cc_library(
# Libraries with GPU facilities that are useful for writing kernels.
cc_library(
name = "gpu_lib",
srcs = if_not_windows([
srcs = [
"common_runtime/gpu/gpu_event_mgr.cc",
]),
hdrs = if_not_windows([
],
hdrs = [
"common_runtime/gpu/gpu_event_mgr.h",
]),
],
copts = tf_copts(),
visibility = ["//visibility:public"],
deps = [
@ -964,7 +964,8 @@ cc_library(
":lib_internal",
":proto_text",
":protos_all_cc",
] + if_not_windows([":stream_executor"]),
":stream_executor",
],
)
cc_library(
@ -982,7 +983,7 @@ tf_proto_library_cc(
name = "worker_proto",
srcs = ["protobuf/worker.proto"],
cc_api_version = 2,
cc_libs = [":protos_all_cc"],
protodeps = [":protos_all"],
visibility = [
"//tensorflow:internal",
],
@ -993,8 +994,8 @@ tf_proto_library_cc(
srcs = ["protobuf/worker_service.proto"],
has_services = 1,
cc_api_version = 2,
cc_libs = [":worker_proto_cc"],
cc_stubby_versions = ["2"],
protodeps = [":worker_proto"],
visibility = [
"//tensorflow:internal",
],
@ -1004,7 +1005,7 @@ tf_proto_library_cc(
name = "master_proto",
srcs = ["protobuf/master.proto"],
cc_api_version = 2,
cc_libs = [":protos_all_cc"],
protodeps = [":protos_all"],
visibility = [
"//tensorflow:internal",
],
@ -1015,8 +1016,8 @@ tf_proto_library_cc(
srcs = ["protobuf/master_service.proto"],
has_services = 1,
cc_api_version = 2,
cc_libs = [":master_proto_cc"],
cc_stubby_versions = ["2"],
protodeps = [":master_proto"],
visibility = [
"//tensorflow:internal",
],
@ -1417,7 +1418,7 @@ tf_cuda_library(
tf_cuda_library(
name = "gpu_runtime",
srcs = if_not_windows([
srcs = [
"common_runtime/gpu/gpu_bfc_allocator.cc",
"common_runtime/gpu/gpu_debug_allocator.cc",
"common_runtime/gpu/gpu_device.cc",
@ -1429,8 +1430,8 @@ tf_cuda_library(
"common_runtime/gpu/pool_allocator.cc",
"common_runtime/gpu/process_state.cc",
"common_runtime/gpu_device_context.h",
]),
hdrs = if_not_windows([
],
hdrs = [
"common_runtime/gpu/gpu_bfc_allocator.h",
"common_runtime/gpu/gpu_debug_allocator.h",
"common_runtime/gpu/gpu_device.h",
@ -1439,7 +1440,7 @@ tf_cuda_library(
"common_runtime/gpu/gpu_util.h",
"common_runtime/gpu/pool_allocator.h",
"common_runtime/gpu/process_state.h",
]),
],
copts = tf_copts(),
linkstatic = 1,
deps = [
@ -1451,10 +1452,9 @@ tf_cuda_library(
":lib",
":lib_internal",
":protos_all_cc",
"//third_party/eigen3",
] + if_not_windows([
":stream_executor",
]),
"//third_party/eigen3",
],
alwayslink = 1,
)

8
tensorflow/core/common_runtime/direct_session_test.cc
View File

@ -835,7 +835,7 @@ static void TestSessionInterOpThreadsImpl(bool use_function_lib) {
FunctionLibraryDefinition flib(OpRegistry::Global(), library_graph_def);
Graph g(&flib);
Tensor t(DT_FLOAT, TensorShape({}));
t.scalar<float>()() = {1.2};
t.scalar<float>()() = {1.2f};
Node* x = test::graph::Constant(&g, t);
Node* y;
if (use_function_lib) {
@ -945,7 +945,7 @@ TEST(DirectSessionTest, TestSessionInterOpThreadsWithFunctions) {
TEST(DirectSessionTest, TestSessionInterOpThreadsInvalidOptions) {
Graph g(OpRegistry::Global());
Tensor t(DT_FLOAT, TensorShape({}));
t.scalar<float>()() = {1.2};
t.scalar<float>()() = {1.2f};
Node* x = test::graph::Constant(&g, t);
GraphDef def;
test::graph::ToGraphDef(&g, &def);
@ -979,7 +979,7 @@ TEST(DirectSessionTest, TestDirectSessionRunClose) {
// Construct a graph with a variable and a single assign.
Graph g(OpRegistry::Global());
Tensor t(DT_FLOAT, TensorShape({}));
t.scalar<float>()() = {1.2};
t.scalar<float>()() = {1.2f};
Node* var_val = test::graph::Constant(&g, t);
Node* var = test::graph::Var(&g, DT_FLOAT, {});
Node* var_assign = test::graph::Assign(&g, var, var_val);
@ -1063,7 +1063,7 @@ TEST(DirectSessionTest, TestDirectSessionReset) {
// Construct a graph with a variable and a single assign.
Graph g(OpRegistry::Global());
Tensor t(DT_FLOAT, TensorShape({}));
t.scalar<float>()() = {1.2};
t.scalar<float>()() = {1.2f};
Node* var_val = test::graph::Constant(&g, t);
Node* var = test::graph::Var(&g, DT_FLOAT, {});
Node* var_assign = test::graph::Assign(&g, var, var_val);

2
tensorflow/core/common_runtime/executor.h
View File

@ -39,7 +39,7 @@ class StepStatsCollector;
// Rendezvous* rendezvous = NewNaiveRendezvous();
// TF_CHECK_OK(rendezvous->Send("input", some_input_tensor));
// TF_CHECK_OK(executor->Run({ExecutorOpts, rendezvous, nullptr}));
// TF_CHECK_OK(rendezvous->Recv("input", &output_tensor));
// TF_CHECK_OK(rendezvous->Recv("output", &output_tensor));
// ... ...
//
// Multiple threads can call Executor::Run concurrently.

14
tensorflow/core/common_runtime/sycl/sycl_allocator.cc
View File

@ -19,16 +19,26 @@ limitations under the License.
namespace tensorflow {
SYCLAllocator::~SYCLAllocator() { }
SYCLAllocator::~SYCLAllocator() {}
string SYCLAllocator::Name() { return "device:SYCL"; }
void *SYCLAllocator::AllocateRaw(size_t alignment, size_t num_bytes) {
assert(device_);
auto p = device_->allocate(num_bytes);
return p;
}
void SYCLAllocator::DeallocateRaw(void *ptr) { device_->deallocate(ptr); }
void SYCLAllocator::DeallocateRaw(void *ptr) {
if (device_) {
device_->deallocate(ptr);
}
}
void SYCLAllocator::EnterLameDuckMode() {
device_->deallocate_all();
device_ = nullptr;
}
} // namespace tensorflow

6
tensorflow/core/common_runtime/sycl/sycl_allocator.h
View File

@ -29,14 +29,16 @@ namespace tensorflow {
class SYCLAllocator : public Allocator {
public:
SYCLAllocator(Eigen::SyclDevice* device) : device_(device) {}
SYCLAllocator(Eigen::QueueInterface* device) : device_(device) {}
virtual ~SYCLAllocator() override;
string Name() override;
void *AllocateRaw(size_t alignment, size_t num_bytes) override;
void DeallocateRaw(void *ptr) override;
void EnterLameDuckMode();
virtual bool ShouldAllocateEmptyTensors() override final { return true; }
private:
Eigen::SyclDevice *device_; // not owned
Eigen::QueueInterface *device_; // not owned
TF_DISALLOW_COPY_AND_ASSIGN(SYCLAllocator);
};

15
tensorflow/core/common_runtime/sycl/sycl_device.cc
View File

@ -25,8 +25,9 @@ namespace tensorflow {
SYCLDevice::~SYCLDevice() {
device_context_->Unref();
delete sycl_allocator_;
sycl_allocator_->EnterLameDuckMode();
delete sycl_device_;
delete sycl_queue_;
}
void SYCLDevice::Compute(OpKernel *op_kernel, OpKernelContext *context) {
@ -50,12 +51,8 @@ Allocator *SYCLDevice::GetAllocator(AllocatorAttributes attr) {
Status SYCLDevice::MakeTensorFromProto(const TensorProto &tensor_proto,
const AllocatorAttributes alloc_attrs,
Tensor *tensor) {
AllocatorAttributes attr;
attr.set_on_host(true);
attr.set_gpu_compatible(true);
Allocator *host_alloc = GetAllocator(attr);
Tensor parsed(tensor_proto.dtype());
if (!parsed.FromProto(host_alloc, tensor_proto)) {
if (!parsed.FromProto(cpu_allocator_, tensor_proto)) {
return errors::InvalidArgument("Cannot parse tensor from proto: ",
tensor_proto.DebugString());
}
@ -86,6 +83,12 @@ Status SYCLDevice::FillContextMap(const Graph *graph,
return Status::OK();
}
Status SYCLDevice::Sync() {
sycl_device_->synchronize();
return Status::OK();
}
} // namespace tensorflow
#endif // TENSORFLOW_USE_SYCL

16
tensorflow/core/common_runtime/sycl/sycl_device.h
View File

@ -22,7 +22,6 @@ limitations under the License.
#define EIGEN_USE_SYCL
#include "tensorflow/core/common_runtime/device_factory.h"
#include "tensorflow/core/common_runtime/local_device.h"
#include "tensorflow/core/common_runtime/sycl/sycl_allocator.h"
#include "tensorflow/core/common_runtime/sycl/sycl_device_context.h"
@ -30,7 +29,6 @@ limitations under the License.
namespace tensorflow {
class SYCLDevice : public LocalDevice {
public:
template <typename SYCLSelector>
@ -42,8 +40,9 @@ public:
name, DEVICE_SYCL, memory_limit, locality,
physical_device_desc), nullptr),
cpu_allocator_(cpu_allocator),
sycl_device_(new Eigen::SyclDevice(sycl_selector)),
sycl_allocator_(new SYCLAllocator(sycl_device_)),
sycl_queue_(new Eigen::QueueInterface(sycl_selector)),
sycl_device_(new Eigen::SyclDevice(sycl_queue_)),
sycl_allocator_(new SYCLAllocator(sycl_queue_)),
device_context_(new SYCLDeviceContext()) {
set_eigen_sycl_device(sycl_device_);
}
@ -59,16 +58,17 @@ public:
Status FillContextMap(const Graph *graph,
DeviceContextMap *device_context_map) override;
Status Sync() override { return Status::OK(); }
Status Sync() override;
static string GetShortDeviceDescription(/*int device_id,
const DeviceDescription& desc*/) {
return strings::StrCat("device: 0, name SYCL, pci bus id: 0");
}
private:
Allocator *cpu_allocator_; // owned
Eigen::SyclDevice* sycl_device_; // owned
SYCLAllocator *sycl_allocator_; // owned
Allocator *cpu_allocator_; // owned
Eigen::QueueInterface* sycl_queue_; // owned
Eigen::SyclDevice* sycl_device_; // owned
SYCLAllocator *sycl_allocator_; // owned
SYCLDeviceContext *device_context_;
};

5
tensorflow/core/common_runtime/sycl/sycl_device_context.cc
View File

@ -16,13 +16,11 @@ limitations under the License.
#if TENSORFLOW_USE_SYCL
#define EIGEN_USE_SYCL
#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
#include "tensorflow/core/common_runtime/sycl/sycl_device_context.h"
#include "tensorflow/core/common_runtime/dma_helper.h"
#define EIGEN_USE_SYCL
#include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor"
namespace tensorflow {
void SYCLDeviceContext::CopyCPUTensorToDevice(const Tensor *cpu_tensor,
@ -108,7 +106,6 @@ void SYCLDeviceContext::CopyDeviceTensorToCPU(const Tensor *device_tensor,
StatusCallback done) {
const int64 total_bytes = device_tensor->TotalBytes();
if (total_bytes > 0) {
device->eigen_sycl_device()->deallocate_all();
const void* src_ptr = DMAHelper::base(device_tensor);
void* dst_ptr = DMAHelper::base(cpu_tensor);
switch (device_tensor->dtype()) {

1
tensorflow/core/common_runtime/sycl/sycl_device_factory.cc
View File

@ -15,6 +15,7 @@ limitations under the License.
#if TENSORFLOW_USE_SYCL
#include "tensorflow/core/common_runtime/device_factory.h"
#include "tensorflow/core/common_runtime/sycl/sycl_device.h"
namespace tensorflow {

2
tensorflow/core/debug/BUILD
View File

@ -36,7 +36,7 @@ tf_proto_library_cc(
has_services = 1,
cc_api_version = 2,
cc_grpc_version = 1,
cc_libs = ["//tensorflow/core:protos_all_cc"],
protodeps = ["//tensorflow/core:protos_all"],
)
# Depending on this target causes a concrete DebuggerState implementation

6
tensorflow/core/debug/debug_gateway_test.cc
View File

@ -372,9 +372,9 @@ TEST_F(SessionDebugMinusAXTest,
debug_gateway.SetNodeValueCallback(
[this, &mu, &val_callback_count, &a_debug_identity_node_name,
&x_debug_identity_node_name, &y_debug_identity_node_name,
&debug_identity_tensor_vals,
&callbacks_done](const string& node_name, const int output_slot,
const Tensor& tensor_value, const bool is_ref) {
&debug_identity_tensor_vals, &callbacks_done, &kConcurrentRuns](
const string& node_name, const int output_slot,
const Tensor& tensor_value, const bool is_ref) {
mutex_lock l(mu);
if (node_name == a_debug_identity_node_name && output_slot == 0) {

6
tensorflow/core/debug/debug_io_utils.cc
View File

@ -18,6 +18,12 @@ limitations under the License.
#include <vector>
#include "grpc++/create_channel.h"
#if defined(PLATFORM_WINDOWS)
// winsock2.h is used in grpc, so Ws2_32.lib is needed
#pragma comment(lib,"Ws2_32.lib")
#endif
#include "tensorflow/core/debug/debug_service.grpc.pb.h"
#include "tensorflow/core/framework/summary.pb.h"
#include "tensorflow/core/lib/io/path.h"

3
tensorflow/core/debug/debug_io_utils_test.cc
View File

@ -273,7 +273,8 @@ TEST_F(DebugIOUtilsTest, PublishTensorConcurrentlyToPartiallyOverlappingPaths) {
auto fn = [this, &dump_count, &done_count, &mu, &dump_root_base, &dump_roots,
&dump_file_paths, &wall_time, &tensor_name, &debug_node_name,
&kNodeName, &kDebugOpName, &kConcurrentPubs, &all_done]() {
&kNodeName, &kDebugOpName, &kConcurrentPubs, &kOutputSlot,
&all_done]() {
// "gumpy" is the shared directory part of the path.
string dump_root;
string debug_url;

2
tensorflow/core/framework/partial_tensor_shape_test.cc
View File

@ -220,7 +220,7 @@ TEST(PartialTensorShapeTest, PartialShapeMergeWith) {
TEST(PartialTensorShapeTest, MakePartialShapeEmpty) {
// Empty made partial shapes should still be fully defined
const int64 dims[0] = {};
const int64 dims[1] = {};
PartialTensorShape shape;
EXPECT_FALSE(shape.IsFullyDefined());
TF_ASSERT_OK(PartialTensorShape::MakePartialShape(dims, 0, &shape));

2
tensorflow/core/framework/tensor_testutil.h
View File

@ -16,6 +16,8 @@ limitations under the License.
#ifndef TENSORFLOW_FRAMEWORK_TENSOR_TESTUTIL_H_
#define TENSORFLOW_FRAMEWORK_TENSOR_TESTUTIL_H_
#include <numeric>
#include "tensorflow/core/framework/tensor.h"
#include "tensorflow/core/lib/gtl/array_slice.h"
#include "tensorflow/core/platform/logging.h"

6
tensorflow/core/kernels/BUILD
View File

@ -2342,7 +2342,6 @@ cc_library(
":batch_norm_op",
":bias_op",
":conv_ops",
":depthwise_conv_grad_op",
":dilation_ops",
":fused_batch_norm_op",
":in_topk_op",
@ -2354,7 +2353,10 @@ cc_library(
":softsign_op",
":topk_op",
":xent_op",
] + if_not_windows([":depthwise_conv_op"]),
] + if_not_windows([
":depthwise_conv_grad_op",
":depthwise_conv_op",
]),
)
NN_DEPS = [

4
tensorflow/core/kernels/adjust_contrast_op_test.cc
View File

@ -56,7 +56,7 @@ TEST_F(AdjustContrastOpTest, Simple_1223) {
TF_EXPECT_OK(InitOp());
AddInputFromArray<float>(TensorShape({1, 2, 2, 3}),
{1, 5, 9, 2, 6, 10, 3, 7, 11, 4, 8, 12});
AddInputFromArray<float>(TensorShape({}), {0.2});
AddInputFromArray<float>(TensorShape({}), {0.2f});
TF_ASSERT_OK(RunOpKernel());
Tensor expected(allocator(), DT_FLOAT, TensorShape({1, 2, 2, 3}));
@ -78,7 +78,7 @@ TEST_F(AdjustContrastOpTest, Big_99x99x3) {
}
AddInputFromArray<float>(TensorShape({1, 99, 99, 3}), values);
AddInputFromArray<float>(TensorShape({}), {0.2});
AddInputFromArray<float>(TensorShape({}), {0.2f});
TF_ASSERT_OK(RunOpKernel());
}

10
tensorflow/core/kernels/batch_norm_op_test.cc
View File

@ -47,15 +47,15 @@ TEST_F(BatchNormOpTest, Simple) {
AddInputFromArray<float>(TensorShape({1, 1, 6, 2}),
{1, 4, 2, 5, 3, 6, -1, -4, -2, -5, -3, -6});
AddInputFromArray<float>(TensorShape({2}), {10, 20});
AddInputFromArray<float>(TensorShape({2}), {0.25, 0.5});
AddInputFromArray<float>(TensorShape({2}), {0.1, 0.6});
AddInputFromArray<float>(TensorShape({2}), {0.0, 0.0});
AddInputFromArray<float>(TensorShape({2}), {0.25f, 0.5f});
AddInputFromArray<float>(TensorShape({2}), {0.1f, 0.6f});
AddInputFromArray<float>(TensorShape({2}), {0.0f, 0.0f});
TF_ASSERT_OK(RunOpKernel());
Tensor expected(allocator(), DT_FLOAT, TensorShape({1, 1, 6, 2}));
test::FillValues<float>(
&expected, {-17.86, -22.00, -15.87, -20.59, -13.87, -19.18, -21.86,
-33.31, -23.85, -34.72, -25.85, -36.13});
&expected, {-17.86f, -22.00f, -15.87f, -20.59f, -13.87f, -19.18f, -21.86f,
-33.31f, -23.85f, -34.72f, -25.85f, -36.13f });
test::ExpectTensorNear<float>(expected, *GetOutput(0), 0.01);
}

15
tensorflow/core/kernels/cast_op_test.cc
View File

@ -49,17 +49,18 @@ class CastOpTest : public OpsTestBase {
TF_EXPECT_OK(InitOp());
}
template <typename IN, typename OUT>
template <typename INPUT, typename OUTPUT>
void CheckCast() {
DataType in_type = DataTypeToEnum<IN>::v();
DataType out_type = DataTypeToEnum<OUT>::v();
DataType in_type = DataTypeToEnum<INPUT>::v();
DataType out_type = DataTypeToEnum<OUTPUT>::v();
MakeOp(in_type, out_type);
AddInputFromArray<IN>(TensorShape({1, 2, 2, 1}),
{IN(1), IN(2), IN(3), IN(4)});
AddInputFromArray<INPUT>(TensorShape({1, 2, 2, 1}),
{INPUT(1), INPUT(2), INPUT(3), INPUT(4)});
TF_ASSERT_OK(RunOpKernel());
Tensor expected(allocator(), out_type, TensorShape({1, 2, 2, 1}));
test::FillValues<OUT>(&expected, {OUT(1), OUT(2), OUT(3), OUT(4)});
test::ExpectTensorEqual<OUT>(expected, *GetOutput(0));
test::FillValues<OUTPUT>(&expected,
{OUTPUT(1), OUTPUT(2), OUTPUT(3), OUTPUT(4)});
test::ExpectTensorEqual<OUTPUT>(expected, *GetOutput(0));
}
};

10
tensorflow/core/kernels/colorspace_op_test.cc
View File

@ -71,7 +71,7 @@ class RGBToHSVOpTest : public OpsTestBase {
void CheckRedMax(DataType data_type) {
// Test case where red channel dominates
AddInputFromArray<T>(TensorShape({3}), {.8, .4, .2});
AddInputFromArray<T>(TensorShape({3}), {.8f, .4f, .2f});
TF_ASSERT_OK(RunOpKernel());
T expected_h = 1. / 6. * .2 / .6;
@ -85,7 +85,7 @@ class RGBToHSVOpTest : public OpsTestBase {
void CheckGreenMax(DataType data_type) {
// Test case where green channel dominates
AddInputFromArray<T>(TensorShape({3}), {.2, .8, .4});
AddInputFromArray<T>(TensorShape({3}), {.2f, .8f, .4f});
TF_ASSERT_OK(RunOpKernel());
T expected_h = 1. / 6. * (2.0 + (.2 / .6));
@ -99,7 +99,7 @@ class RGBToHSVOpTest : public OpsTestBase {
void CheckBlueMax(DataType data_type) {
// Test case where blue channel dominates
AddInputFromArray<T>(TensorShape({3}), {.4, .2, .8});
AddInputFromArray<T>(TensorShape({3}), {.4f, .2f, .8f});
TF_ASSERT_OK(RunOpKernel());
T expected_h = 1. / 6. * (4.0 + (.2 / .6));
@ -112,7 +112,7 @@ class RGBToHSVOpTest : public OpsTestBase {
}
void CheckNegativeDifference(DataType data_type) {
AddInputFromArray<T>(TensorShape({3}), {0, .1, .2});
AddInputFromArray<T>(TensorShape({3}), {0, .1f, .2f});
TF_ASSERT_OK(RunOpKernel());
T expected_h = 1. / 6. * (4.0 + (-.1 / .2));
@ -220,7 +220,7 @@ class HSVToRGBOpTest : public OpsTestBase {
TF_ASSERT_OK(RunOpKernel());
Tensor expected(allocator(), data_type, TensorShape({3}));
test::FillValues<T>(&expected, {0, .1, .2});
test::FillValues<T>(&expected, {0, .1f, .2f});
test::ExpectTensorNear<T>(expected, *GetOutput(0), 1e-6);
}
};

26
tensorflow/core/kernels/control_flow_ops.cc
View File

@ -113,9 +113,12 @@ REGISTER_GPU_HOST_REF_KERNEL(string);
#undef REGISTER_GPU_HOST_REF_KERNEL
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER( \
Name("Switch").Device(DEVICE_SYCL).TypeConstraint<type>("T"), SwitchOp)
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("Switch") \
.Device(DEVICE_SYCL) \
.TypeConstraint<type>("T") \
.HostMemory("pred"), \
SwitchOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#undef REGISTER_SYCL_KERNEL
@ -219,9 +222,12 @@ REGISTER_GPU_REF_KERNEL(bool);
#undef REGISTER_GPU_REF_KERNEL
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER( \
Name("Merge").Device(DEVICE_SYCL).TypeConstraint<type>("T"), MergeOp)
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER(Name("Merge") \
.Device(DEVICE_SYCL) \
.TypeConstraint<type>("T") \
.HostMemory("value_index"), \
MergeOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#undef REGISTER_SYCL_KERNEL
@ -418,8 +424,12 @@ REGISTER_GPU_HOST_KERNEL(string);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(type) \
REGISTER_KERNEL_BUILDER( \
Name("NextIteration").Device(DEVICE_SYCL).TypeConstraint<type>("T"), NextIterationOp)
REGISTER_KERNEL_BUILDER(Name("NextIteration") \
.Device(DEVICE_SYCL) \
.HostMemory("data") \
.HostMemory("output") \
.TypeConstraint<type>("T"), \
NextIterationOp)
REGISTER_SYCL_KERNEL(bool);
TF_CALL_NUMBER_TYPES_NO_INT32(REGISTER_SYCL_KERNEL);
#undef REGISTER_SYCL_KERNEL

18
tensorflow/core/kernels/control_flow_ops_test.cc
View File

@ -85,13 +85,27 @@ class AbortOpTest : public OpsTestBase {
protected:
};
#ifdef PLATFORM_WINDOWS
#define SIGABRT 3
class KilledBySignal {
public:
explicit KilledBySignal(int signum) : signum_(signum) {}
bool operator()(int exit_status) const { return exit_status == signum_; }
private:
const int signum_;
};
#else
#define KilledBySignal ::testing::KilledBySignal
#endif
// Pass an error message to the op.
TEST_F(AbortOpTest, pass_error_msg) {
TF_ASSERT_OK(NodeDefBuilder("abort_op", "Abort")
.Attr("error_msg", "abort_op_test")
.Finalize(node_def()));
TF_ASSERT_OK(InitOp());
EXPECT_EXIT(RunOpKernel(), ::testing::KilledBySignal(SIGABRT),
EXPECT_EXIT(RunOpKernel(), KilledBySignal(SIGABRT),
"Abort_op intentional failure; abort_op_test");
}
@ -99,7 +113,7 @@ TEST_F(AbortOpTest, pass_error_msg) {
TEST_F(AbortOpTest, default_msg) {
TF_ASSERT_OK(NodeDefBuilder("abort_op", "Abort").Finalize(node_def()));
TF_ASSERT_OK(InitOp());
EXPECT_EXIT(RunOpKernel(), ::testing::KilledBySignal(SIGABRT),
EXPECT_EXIT(RunOpKernel(), KilledBySignal(SIGABRT),
"Abort_op intentional failure; ");
}

12
tensorflow/core/kernels/cwise_op_abs.cc
View File

@ -21,6 +21,18 @@ REGISTER5(UnaryOp, CPU, "Abs", functor::abs, float, Eigen::half, double, int32,
#if !defined(IS_MOBILE_PLATFORM)
REGISTER2(UnaryOp, CPU, "ComplexAbs", functor::abs, complex64, complex128);
#endif
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Abs") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::abs<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER4(UnaryOp, GPU, "Abs", functor::abs, float, Eigen::half, double, int64);
REGISTER2(UnaryOp, GPU, "ComplexAbs", functor::abs, complex64, complex128);

12
tensorflow/core/kernels/cwise_op_acos.cc
View File

@ -17,6 +17,18 @@ limitations under the License.
namespace tensorflow {
REGISTER2(UnaryOp, CPU, "Acos", functor::acos, float, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Acos") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::acos<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER2(UnaryOp, GPU, "Acos", functor::acos, float, double);
#endif

2
tensorflow/core/kernels/cwise_op_add_1.cc
View File

@ -26,7 +26,7 @@ REGISTER5(BinaryOp, CPU, "Add", functor::add, float, Eigen::half, double, int32,
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
BinaryOp<SYCLDevice, functor::add<TYPE>>);
TF_CALL_NUMBER_TYPES(REGISTER_SYCL_KERNEL);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL

12
tensorflow/core/kernels/cwise_op_asin.cc
View File

@ -17,6 +17,18 @@ limitations under the License.
namespace tensorflow {
REGISTER2(UnaryOp, CPU, "Asin", functor::asin, float, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Asin") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::asin<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER2(UnaryOp, GPU, "Asin", functor::asin, float, double);
#endif

12
tensorflow/core/kernels/cwise_op_atan.cc
View File

@ -17,6 +17,18 @@ limitations under the License.
namespace tensorflow {
REGISTER2(UnaryOp, CPU, "Atan", functor::atan, float, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Atan") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::atan<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER2(UnaryOp, GPU, "Atan", functor::atan, float, double);
#endif

12
tensorflow/core/kernels/cwise_op_ceil.cc
View File

@ -17,6 +17,18 @@ limitations under the License.
namespace tensorflow {
REGISTER3(UnaryOp, CPU, "Ceil", functor::ceil, float, Eigen::half, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Ceil") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::ceil<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Ceil", functor::ceil, float, Eigen::half, double);
#endif

12
tensorflow/core/kernels/cwise_op_cos.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER5(UnaryOp, CPU, "Cos", functor::cos, float, Eigen::half, double,
complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Cos") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::cos<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Cos", functor::cos, float, Eigen::half, double);
#endif

5
tensorflow/core/kernels/cwise_op_div.cc
View File

@ -30,6 +30,11 @@ REGISTER5(BinaryOp, CPU, "RealDiv", functor::div, float, Eigen::half, double,
Name("Div") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
BinaryOp<SYCLDevice, functor::div<TYPE>>); \
REGISTER_KERNEL_BUILDER( \
Name("RealDiv") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
BinaryOp<SYCLDevice, functor::div<TYPE>>);
REGISTER_SYCL_KERNEL(float)
#undef REGISTER_SYCL_KERNEL

12
tensorflow/core/kernels/cwise_op_exp.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER5(UnaryOp, CPU, "Exp", functor::exp, float, Eigen::half, double,
complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Exp") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::exp<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Exp", functor::exp, float, Eigen::half, double);
#endif

12
tensorflow/core/kernels/cwise_op_floor.cc
View File

@ -17,6 +17,18 @@ limitations under the License.
namespace tensorflow {
REGISTER3(UnaryOp, CPU, "Floor", functor::floor, float, Eigen::half, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Floor") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::floor<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Floor", functor::floor, float, Eigen::half, double);
#endif

8
tensorflow/core/kernels/cwise_op_floor_div.cc
View File

@ -18,6 +18,9 @@ limitations under the License.
namespace tensorflow {
REGISTER5(BinaryOp, CPU, "FloorDiv", functor::safe_floor_div, uint8, uint16,
int16, int32, int64);
REGISTER3(BinaryOp, CPU, "FloorDiv", functor::floor_div_real, float,
Eigen::half, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
@ -25,11 +28,10 @@ REGISTER5(BinaryOp, CPU, "FloorDiv", functor::safe_floor_div, uint8, uint16,
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
BinaryOp<SYCLDevice, functor::floor_div<TYPE>>);
TF_CALL_INTEGRAL_TYPES(REGISTER_SYCL_KERNEL);
REGISTER_SYCL_KERNEL(float)
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
REGISTER3(BinaryOp, CPU, "FloorDiv", functor::floor_div_real, float,
Eigen::half, double);
#if GOOGLE_CUDA
REGISTER4(BinaryOp, GPU, "FloorDiv", functor::floor_div, uint8, uint16, int16,
int64);

4
tensorflow/core/kernels/cwise_op_isfinite.cc
View File

@ -18,6 +18,7 @@ limitations under the License.
namespace tensorflow {
REGISTER3(UnaryOp, CPU, "IsFinite", functor::isfinite, float, Eigen::half,
double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
@ -25,9 +26,10 @@ REGISTER3(UnaryOp, CPU, "IsFinite", functor::isfinite, float, Eigen::half,
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::isfinite<TYPE>>);
TF_CALL_REAL_NUMBER_TYPES(REGISTER_SYCL_KERNEL);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "IsFinite", functor::isfinite, float, Eigen::half,
double);

4
tensorflow/core/kernels/cwise_op_isinf.cc
View File

@ -17,6 +17,7 @@ limitations under the License.
namespace tensorflow {
REGISTER3(UnaryOp, CPU, "IsInf", functor::isinf, float, Eigen::half, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
@ -24,9 +25,10 @@ REGISTER3(UnaryOp, CPU, "IsInf", functor::isinf, float, Eigen::half, double);
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::isinf<TYPE>>);
TF_CALL_REAL_NUMBER_TYPES(REGISTER_SYCL_KERNEL);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "IsInf", functor::isinf, float, Eigen::half, double);
#endif

4
tensorflow/core/kernels/cwise_op_isnan.cc
View File

@ -17,6 +17,7 @@ limitations under the License.
namespace tensorflow {
REGISTER3(UnaryOp, CPU, "IsNan", functor::isnan, float, Eigen::half, double);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
@ -24,9 +25,10 @@ REGISTER3(UnaryOp, CPU, "IsNan", functor::isnan, float, Eigen::half, double);
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::isnan<TYPE>>);
TF_CALL_REAL_NUMBER_TYPES(REGISTER_SYCL_KERNEL);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "IsNan", functor::isnan, float, Eigen::half, double);
#endif

12
tensorflow/core/kernels/cwise_op_log.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER5(UnaryOp, CPU, "Log", functor::log, float, Eigen::half, double,
complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Log") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::log<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Log", functor::log, float, Eigen::half, double);
#endif

12
tensorflow/core/kernels/cwise_op_log1p.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER5(UnaryOp, CPU, "Log1p", functor::log1p, float, Eigen::half, double,
complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Log1p") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::log1p<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Log1p", functor::log1p, float, Eigen::half, double);
#endif

12
tensorflow/core/kernels/cwise_op_neg.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER7(UnaryOp, CPU, "Neg", functor::neg, float, Eigen::half, double, int32,
complex64, int64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Neg") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::neg<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER4(UnaryOp, GPU, "Neg", functor::neg, float, Eigen::half, double, int64);

12
tensorflow/core/kernels/cwise_op_pow.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER7(BinaryOp, CPU, "Pow", functor::pow, float, Eigen::half, double, int32,
int64, complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Pow") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
BinaryOp<SYCLDevice, functor::pow<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER4(BinaryOp, GPU, "Pow", functor::pow, float, Eigen::half, double,
int64);

12
tensorflow/core/kernels/cwise_op_rsqrt.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER5(UnaryOp, CPU, "Rsqrt", functor::rsqrt, float, Eigen::half, double,
complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Rsqrt") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::rsqrt<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYCL
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Rsqrt", functor::rsqrt, float, Eigen::half, double);
#endif

12
tensorflow/core/kernels/cwise_op_sin.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER5(UnaryOp, CPU, "Sin", functor::sin, float, Eigen::half, double,
complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Sin") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::sin<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYC
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Sin", functor::sin, float, Eigen::half, double);
#endif

12
tensorflow/core/kernels/cwise_op_sqrt.cc
View File

@ -18,6 +18,18 @@ limitations under the License.
namespace tensorflow {
REGISTER5(UnaryOp, CPU, "Sqrt", functor::sqrt, float, Eigen::half, double,
complex64, complex128);
#if TENSORFLOW_USE_SYCL
#define REGISTER_SYCL_KERNEL(TYPE) \
REGISTER_KERNEL_BUILDER( \
Name("Sqrt") \
.Device(DEVICE_SYCL) \
.TypeConstraint<TYPE>("T"), \
UnaryOp<SYCLDevice, functor::sqrt<TYPE>>);
REGISTER_SYCL_KERNEL(float);
#undef REGISTER_SYCL_KERNEL
#endif // TENSORFLOW_USE_SYC
#if GOOGLE_CUDA
REGISTER3(UnaryOp, GPU, "Sqrt", functor::sqrt, float, Eigen::half, double);
#endif

Some files were not shown because too many files have changed in this diff Show More