TensorFlow Community keeps rolling with updates. The first release candidate for next minor version release 1.9.0 is unveiled today with pretty good list of features, improvements and bug fixes. In its previous version 1.8.0 release, the team paid more attention towards supporting GPU memory, running on multiple GPUs and cloud performance. In today’s release, the team were strong in adding support to Keras, gradient estimators and improvement in the layers.
Major features and improvements in TensorFlow 1.9.0-rc0:
- Updated tf.keras to the Keras 2.1.6 API.
- tfe.Network is deprecated and can be inherited from tf.keras.Model.
- Added support of core feature columns and losses to gradient boosted trees estimators.
- The distributions.Bijector API supports broadcasting for Bijectors with new API changes.
- Layered variable names changed
Bug Fixes in TensorFlow 1.9.0-rc0:
- The DatasetBase::DebugString() method is now const.
- Added the tf.contrib.data.sample_from_datasets() API for randomly sampling from multiple datasets.
- Eager Execution and Accelerated Linear Algebra (XLA) fixed.
- tf.keras.Model.save_weights by default saves in TensorFlow format.
- TensorFlow Debugger (tfdbg) CLI fixed.
- Added “constrained_optimization” to tensorflow/contrib.
- tf.contrib.framework.zero_initializer supports ResourceVariable.
- tf.contrib.data.make_csv_dataset() supports line breaks in quoted strings.
Miscellaneous changes:
- Added GCS Configuration Ops.
- MakeIterator signature changed to enable propagating error status.
- KL divergence for two Dirichlet distributions.
- More consistent GcsFileSystem behavior for reads past EOF.
- Added Benchmark for tf.scan in graph and eager modes.
- Added complex128 support to FFT, FFT2D, FFT3D, IFFT, IFFT2D, and IFFT3D.
- Support for preventing tf.gradients() from backpropagating through integer tensors.
- Supports indicator column in boosted trees.
- Conv3D, Conv3DBackpropInput, Conv3DBackpropFilter now supports arbitrary.
- LinearOperator[1D,2D,3D]Circulant added to tensorflow.linalg.
- Allows LinearOperator to broadcast.
For the complete list of bug fixes and improvements, you can read TensorFlow’s Github page.
You can also download the source code to access all the exciting features of TensorFlow 1.9.0-rc0.
Read Next
Implementing feedforward networks with TensorFlow
How TFLearn makes building TensorFlow models easier
Distributed TensorFlow: Working with multiple GPUs and servers