* conv3d_tranpose in tf and th
* fix _preprocess_deconv_output_shape error
* cntk conv3d_tranpose
* conv3d_tranpose test
* formatting
* cleanup tests
* fix incorrect axis ordering and docs
* fix incorrect axis ordering and docs
* deconv3d_output_shape to fix errors
* remove conv2d_transpose reference in theano backend
* remove kernel_size loop from test
* put depth first in test and add dim to invalid use case input
* formatting - removed extra line
* fix pep8
* remove extraneous args from tf conv3d_transpose function
* default val for data_format=None
* Add MobileNet to application
* Add support for 1001 classes in imagenet utils
* Revert a mistake in the tests
* Setup application test for mobilenet to run only when on tensorflow
* Correct pytest.mark.skipif explanation for skipping tests if not on tensorflow
* Corrected mobilenet to support 1000 classes and reverted imagenet_utils to prior state
* Fix tensorflow test
* Restrict mobilenets to data format "channels_last"
* Add review fixes
* PEP8 fix
* Add relu6 to activations.py
* Corrected imports in mobilenet.py
* Rolled back activation relu6 and inlined it to mobilenet.py
* Refactored DepthwiseConv2D and other corrections
* Fixed tests
* PEP8 correction
* Add docs to private functions and other fixes
* Fix failed test where input shape is None
* Fix value of size for model name
* Add more numpy-style attributes to HDF5Matrix
* Improve docstrings
* Add test coverage
* Add a ´# Returns´ section to shape, dtype, ndim, size.
* Remove whitespace in blank lines
* Use third-person and close docstrings on a new line.
* Utility function to check if a callable has a given keyword argument
* Added unit tests for the has_arg function
* Replace uses of getargspec with the new has_arg function
Not changing keras.backend, because that gives ImportErrors due to
a circular import (conv_utils uses the backend, and is imported
before generic_utils in utils/__init__.py)
Not changing keras.utils.test_utils, because that change exposes
(what looks to me like) a latent bug
* Replace incorrect use of getargspec in test_utils.py
The previous code would always fail to detect the 'weights' argument.
Simply replacing getargspec would cause the tests for some of the legacy
layers to fail because the passed 'weights' argument is bad.
Instead, I have added a check for whether the passed `weights` array
is empty, this avoids tripping the bug.
* Replacing getargspec with has_arg in the backend modules
This requires reordering imports to avoid errors caused by
conv_utils trying to import the backend, the backend wanting to
import generic_utils, and utils/__init__.py listing conv_utils
before generic_utils.
* Removed getargspec from legacy wrapping function
Instead save the wrapped function in an attribute and call
getargspec on this attribute during documentation generation.
* Initial support for Datasets
* Fix warnings
* Fix for python2
* Fix travis deps
* Fix python2 indexing
* Fix test and docs
* Avoir use of future, use multiprocessing.pool
* Changed warning and better moduling
* fix threading test
* Move Dataset and enqueuers to utils.data_utils
* Skip None input, add seed for generators
* Skip None input fix
* pep8
* Fix example
* Add test in training and changed Dataset to hold item
* Revert to batch handling
* Docs update
* PEP8
* Rename in test
* Better documentation in Sequence
* Typo in sequence warning
* Rename pickle_safe and max_q_size, typos
* Typo in docstring
* Fix tests in training
* Utility function to check if a callable has a given keyword argument
* Replace uses of getargspec with the new has_arg function
Not changing keras.backend, because that gives ImportErrors due to
a circular import (conv_utils uses the backend, and is imported
before generic_utils in utils/__init__.py)
Not changing keras.utils.test_utils, because that change exposes
(what looks to me like) a latent bug
* Added unit tests for the has_arg function
* Replace incorrect use of getargspec in test_utils.py
The previous code would always fail to detect the 'weights' argument.
Simply replacing getargspec would cause the tests for some of the legacy
layers to fail because the passed 'weights' argument is bad.
Instead, I have added a check for whether the passed `weights` array
is empty, this avoids tripping the bug.
* Replacing getargspec with has_arg in the backend modules
This requires reordering imports to avoid errors caused by
conv_utils trying to import the backend, the backend wanting to
import generic_utils, and utils/__init__.py listing conv_utils
before generic_utils.
* Warn always about semantic changes if having keras1 args in *_generator calls.
* modified api upgrade warning message to be more detailed
* minor fix to pep8 syntax
* Use the pytest tmpdir fixture (#6881)
* Run test_data_utils in a temporary directory
* Check output using os.path.isdir or os.path.isfile instead of os.path.exists
* Use the tmpdir fixture instead of mkdtemp
* Use in_tmpdir fixture when writing files in tests
... to avoid leaving files in the repository when tests fail, and also to
avoid the possibility of race conditions when several tests try to access
the same file.
* Serialize/Deserialize numpy arrays passed as arguments to Lambda layers
* Serialize/Deserialize numpy arrays passed as arguments to Lambda layers
* Corrections from fchollet comments
* corrections
* Removes warning and adds a unit test
* pep8 corrections
* Add top_k_sparse_categorical_accuracy and test_top_k_sparse_categorical_accuracy
* Rename top_k_sparse_categorical_accuracy and sparse_top_k_categorical_accuracy
* Better error message for invalid funcational api inputs (#6589)
* raise ValueError if `inputs` is not a Keras tensor
* Move to respective backends
* raise error if is_keras_tensor is called on a non-tensor object
* Fix failing tests
* responding to comments
* Update docstring comments to better explain expected behavior
Taehoon Lee