TensorFlow 1.9.0-rc2 发布，改进数据加载和文本处理

TensorFlow 1.9.0-rc2 已发布。

该版本的更新内容包括新特性的增加、Bug 修复和其他改进，以及一些破坏性的变化。

主要特性和改进

• Updated docs for tf.keras: New Keras-based get started,
  and programmers guide page.

• Update tf.keras to the Keras 2.1.6 API.

• Added tf.keras.layers.CuDNNGRU and tf.keras.layers.CuDNNLSTM layers. Try it.

• Adding support of core feature columns and losses to gradient boosted trees estimators.

• The python interface
  for the TFLite Optimizing Converter
  has been expanded, and the command line interface (AKA: toco, tflite_convert) is once again
  included in the standard pip installation.

• Improved data-loading and text processing with:

• tf.decode_compressed

• tf.string_strip

• tf.strings.regex_full_match

• Added experimental support for new pre-made Estimators:

• tf.contrib.estimator.BaselineEstimator

• tf.contrib.estimator.RNNClassifier

• tf.contrib.estimator.RNNEstimator

• The distributions.Bijector
  API supports broadcasting for Bijectors with new API changes.

破坏性变化

• If you're opening empty variable scopes; replace variable_scope('', ...) by
  variable_scope(tf.get_variable_scope(), ...).

• Headers used for building custom ops have been moved from site-packages/external into site-packages/tensorflow/include/external.

源码下载和详细更新说明请查看 https://github.com/tensorflow/tensorflow/releases/tag/v1.9.0-rc2