Major Features and Improvements
- Adds support for Python 3.5
- Initial version of following orchestration platform supported:
- Kubeflow
- Added TensorFlow Model Analysis Colab example
- Supported split ratio for ExampleGen components
- Supported running a single executor independently
Bug fixes and other changes
- Fixes issue #43 that prevent new execution in some scenarios
- Fixes issue #47 that causes ImportError on chicago_taxi execution on dataflow
- Depends on
apache-beam[gcp]>=2.12,<3 - Depends on
tensorflow-data-validation>=0.13.1,<0.14 - Depends on
tensorflow-model-analysis>=0.13.2,<0.14 - Depends on
tensorflow-transform>=0.13,<0.14 - Deprecations:
- PipelineDecorator is deprecated. Please construct a pipeline directly from a list of components instead.
- Increased verbosity of logging to container stdout when running under
Kubeflow Pipelines. - Updated developer tutorial to support Python 3.5+
Breaking changes
- Examples code are moved from 'examples' to 'tfx/examples': this ensures that PyPi package contains only one top level python module 'tfx'.
Things to notice for upgrading
- Multiprocessing on Mac OS >= 10.13 might crash for Airflow. See
AIRFLOW-3326
for details and solution.