Spark Deep Learning Versions Save

• Deprecated some submodules and classes
• Updated dependencies

• HorovodRunner's main function can now have a return value.
• Better support for multiple GPU machines (Databricks only version).

• Upgrade to TensorFlow 1.12.0.

• Added HorovovodRunner API.
• Simplified test and doc build w/ Docker and conda.
• Updated public Python API docs.
• Removed persistence from DeepImageFeaturizer.

• ignore nullable in DeepImageFeaturizer.validateSchema
• upgrade TensorFrames version to 0.5.0
• upgrade Tensorflow version to 1.10.0 and Keras version to 2.2.2

• keras_image_file_estimator support both sparse and dense vectors
• upgrade TensorFrames version to 0.4.0
• add style checks to Travis CI
• doc fixes

This is the 1.0.0 release. It brings compatibility with newer versions of Spark (2.3) and Tensorflow (1.6+). The custom image schema formerly defined in this package has been replaced with Spark's ImageSchema so there may be some breaking changes when updating to this version.

Notable changes:

• (breaking change) Using the definition of images from Spark 2.3.0. The new definition uses the BGR channel ordering for 3-channel images instead of the RGB ordering used in this project before the change.
• Persistence for DeepImageFeaturizer (both Python and Scala).

This is the final release of dl-pipelines prior to migrating to new ImageSchema.

Notable changes:

• Added vgg16, vgg19 models to DeepImageFeaturizer/DeepImagePredictor (Python).
• Added a Scala API for DeepImageFeaturizer (for transfer learning for images).
• Added TFTransformer and KerasTransformer for applying TensorFlow graphs or TensorFlow-backed Keras models to a column of arrays in a Spark DataFrame.

This is the final release for Deep Learning Pipelines 0.2.0

Notable additions since 0.1.0:

• KerasImageFileEstimator API (train a Keras model on image files)
• SQL UDF support for Keras models
• Added Xception, Resnet50 models to DeepImageFeaturizer/DeepImagePredictor.