| Commit message (Collapse) | Author | Age |
|
|
|
| |
Change: 110422306
|
|
|
|
|
| |
Note: these are not final.
Change: 110420744
|
| |
|
|
|
|
| |
Change: 110417789
|
|
|
|
|
|
| |
accuracy by effectively increasing window size after subsampling.
Change: 110417524
|
| |
|
|
|
|
| |
Change: 110414708
|
| |
|
|
|
|
|
|
| |
unused import.
Change: 110411627
|
|
|
|
| |
Change: 110406666
|
|
|
|
|
|
|
| |
receiver object's converting constructor. This makes it such that the
r.h.s can be an arbitrary C++ expression and is evaluated completely
before getting passed to the constructor.
Change: 110404504
|
|
|
|
|
|
| |
This avoids a CHECK failure if invalid input data is passed to some
ops. It also tightens up the Python class to reject negative dimensions.
Change: 110401558
|
|
|
|
|
|
|
| |
* copy to get .bazelrc and add --verbose_failures
* copy install scripts one by one (it is no longer needed to re-run
all the scripts when only the last one changes)
Change: 110398445
|
|
|
|
|
|
| |
* users can see that there may be some temporary issues
with their platform
Change: 110398280
|
|
|
|
| |
Change: 110397907
|
|
|
|
|
|
| |
* it will leave the jenkins workspace (outside docker) with useful
file permissions and working symbolic links
Change: 110397897
|
|
|
|
|
|
| |
- Calling REGISTER_OP after initialization (initialized_ == true) would
cause OpDefBuilder to be deleted before fields can be added to it.
Change: 110394914
|
|
|
|
| |
Change: 110379526
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The gradient function was previously generating an invalid
IndexedSlices, whereby `IndexedSlices.indices` tensor was not a
vector. This change reshapes the indices and gradient so that they can
correctly be interpreted as an IndexedSlices and applied to the
embedding variable.
Added a multi-dimensional gradient test in embedding_ops_test.py.
Fixes #505. Partially addresses #464.
Change: 110364370
|
|
|
|
|
|
|
| |
1. Fixed default tensor dtypes in the gradients of control flow ops.
2. Better colocation decisions for control flow ops.
3. Don't use ref for loop variables. This would give more flexibility to placement.
Change: 110308978
|
|
|
|
|
|
| |
i.e., does something like "lambda: [x, x]".
Change: 110306609
|
|
|
|
|
|
| |
This should make more shape inference possible, when concat is used to
build, e.g., shape vectors.
Change: 110303876
|
|
|
|
|
| |
Follow-up to 110287597.
Change: 110290160
|
|
|
|
|
|
| |
typescript, and the compiled javascript code should be ignored.
Change: 110288218
|
|
|
|
|
|
| |
compressedHistograms.
Change: 110287931
|
|
|
|
| |
Change: 110287597
|
| |
|
|
|
|
| |
Change: 110286577
|
|
|
|
| |
Change: 110283867
|
|
|
|
| |
Change: 110278005
|
|
|
|
| |
Change: 110277312
|
|
|
|
| |
Change: 110273215
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The repr replaces the default Python representation for an object with
information about the Tensor's name, shape and dtype. TensorShape's
new str() is a more compact representation that makes it easier to
tell, at a glance, what the shape represents.
For example:
print repr(tf.placeholder(tf.qint32, shape=(32, None, 2), name="c"))
# ==> <tf.Tensor 'c:0' shape=(32, ?, 2) dtype=qint32>
Fixes #460.
Change: 110268564
|
|
|
|
|
|
| |
OS X.
Change: 110256863
|
| |
|
|
|
|
| |
Change: 110220692
|
|
|
|
| |
Change: 110208316
|
|
|
|
| |
Change: 110202067
|
|
|
|
| |
Change: 110201665
|
|
|
|
| |
Change: 110191670
|
|
|
|
|
|
| |
This allows one to retrieve QueueRunners corresponding to queues in a specific
scope using tf.get_collection(tf.GraphKeys.QUEUE_RUNNERS, scope).
Change: 110169009
|
|
|
|
| |
Change: 110167188
|
|
|
|
|
|
| |
components are now under `components` directory. This follows polymer conventions.
Change: 110158503
|
|
|
|
| |
Change: 110155285
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Change 110055925
Clean up interface for adjust_contrast and adjust_brightness.
- Simplify kernel for adjust_contrast and remove all min/max and casts.
- Change semantics of delta arg to adjust_brightness (always in [0,1)), and adjust users.
- Add saturate_cast for casting images without over/underflow problems.
- Add new numbers for adjust_contrast benchmark.
This CL makes two changes to the public API:
- It changes the semantics of the delta parameter of adjust_brightness, which was in the same range as the input image before, and now is always in [0,1).
- It changes the semantics of adjust_contrast (the cc op), which wasn't hidden, but was shadowed by the python wrapper in image_ops. It's a little questionable whether this function was part of the public API. It definitely shouldn't have been. It is now hidden, although now it could be part of the public API, albeit with a different name.
Change 110054427
update ci_build
* add PYTHON_BIN_PATH and always run ./configure in ci_build
* rename ci_build cache directory to bazel-ci_build-cache
* sync ci_build/Dockerfile.cpu with docker/Dockerfile.devel
* use "FROM nvidia/cuda:..." for gpu container
* therefore no need of the tensorflow_extra_deps directory anymore
* share install code between containers using ./install/*.sh scripts
* do not inherit (and override FROM clausule in dockerfiles anymore)
* print bazel test errors to stderr
Change 110047126
Update ops.pbtxt.
Change 110046428
Simplify the example for the Fill op.
Base CL: 110056265
|
|
|
|
|
|
|
| |
Change 110044026
Wrap comment in build_pip_package.sh
Base CL: 110044081
|
|
|
|
|
|
|
| |
Change 110024345
Removed the unary operator restriction on MaxPool so we can reuse it more flexibly in the future
Base CL: 110043747
|
|
|
|
| |
Base CL: 110018194
|
|
|
|
|
|
|
|
|
|
|
|
| |
Change 110010103
Implementing SparseSplitOp.
The op takes a sparse tensor (list, values and shape), split_dim and num_splits and produces a list of num_splits tensors where the shape of each tensor is the shape of the original tensor except split_dim = shape[split_dim +num_split - 1 / num_split]. in case if shape[split_dim] is not an integer multiple of num_split an extra one dimension get added to the slices starting from 0.
For example if the input shape is a [2, 10] split_dim = 1, num_split = 3
output shapes will be [[2, 4], [2, 4], [2, 2]].
The Op register shape to [Unknown, dim] for indices tensors and [Unknown] for the values tensor because shape can't be inferred without evaluate input tensors.
Base CL: 110012853
|