v2020.4.29: Multitask

Very similar to v2020.4.23: Multitask apart from a few corrections

• newest results in README
• corrections to output path
• output for return_tokens=False case of split_parse

The model version stays the same.

06/08/2020 addition: deep_reference_parser-2020.8.5-py3-none-any.whl has been added to the binaries. This .whl contains pinned versions of Keras and Tensorflow.

v2020.4.23: Multitask

• Add multitask split_parse command and tests, called with python -m deep_reference_parser split_parse
• Fix issues with training data creation
• Output predictions of validation data by default
• Various improvements - using tox for testing, refactoring, improving error messages, README and tests

v2020.3.1: Prep for Multitask

• load_tsv can now deal with an arbitrary number of labels.
• prodigy_to_tsv now takes a csv list of token labelled prodigy docs as first argument, and will output a single tsv file with combined labels ready for load_tsv
• Sub modules reorganised: all io functions moved to the io module, all functions related to reading and writing prodigy format moved to the prodigy module.
• Linting and other small improvements.

v2020.3.0: Add parse command

• Adds parse command that can be called with python -m deep_reference_parser parse
• Rename predict command to 'split' which can be called with python -m deep_reference_parser parse
• Squashes most tensorflow, keras_contrib, and numpy warnings in __init__.py resulting from old versions and soon-to-be deprecated functions.
• Reduces verbosity of logging, improving CLI clarity.
• Fixes issues in annotate_numbered_references command.

v2020.2.0: Pre-Release

Pre-Release version of deep_reference_parser. Features train and predict functions tested mainly for the task of labelling reference (e.g. academic references) spans in policy documents (e.g. documents produced by government, NGOs, etc).