Release 0.12.0

Major Features and Improvements

• Add support for computing statistics over slices of data.
• Performance improvement due to optimizing inner loops.
• Add support for generating statistics from a pandas dataframe.
• Performance improvement due to pre-allocating tf.Example in
  TFExampleDecoder.
• Performance improvement due to merging common stats generator, numeric stats
  generator and string stats generator as a single basic stats generator.
• Performance improvement due to merging top-k and uniques generators.
• Add a validate_instance function, which checks a single example for
  anomalies.
• Add a utility method get_statistics_html, which returns HTML that can be
  used for Facets visualization outside of a notebook.
• Add support for schema inference of semantic domains.
• Performance improvement on statistics computation over a pandas dataframe.

Bug Fixes and Other Changes

• Use constant 'BYTES_VALUE' in the statistics proto to represent a bytes
  value which cannot be decoded as a utf-8 string.
• Introduced CombinerFeatureStatsGenerator, a specialized interface for
  combiners that do not require cross-feature computations.
• Expand unit test coverage.
• Add optional frequency threshold that allows keeping only the most frequent
  values that are present in a minimum number of examples.
• Add optional desired batch size that allows specification of the number of
  examples to include in each batch.
• Depends on numpy>=1.14.5,<2.
• Depends on protobuf>=3.6.1,<4.
• Depends on apache-beam[gcp]>=2.10,<3.
• Depends on tensorflow-metadata>=0.12.1,<0.13.
• Depends on scikit-learn>=0.18,<1.
• Depends on IPython>=5.0.
• Requires pre-installed tensorflow>=1.12,<2.
• Revise example notebook and update it to be able to run in Colab and Jupyter.

Breaking Changes

• Represent batch as a list of ndarrays instead of ndarrays of ndarrays.
• Modify decoders to return ndarrays of type numpy.float32 for FLOAT features.