indralab/adeft: 0.6.0

In this release

• Users may specify a seed that will be used in random number generators involved in adeft model creation, allowing for repeatable model training results.
• Additional statistics are captured at the time of model training. F1, Precision, and Recall are now captured for each class label separately, allowing users to see how performance compares across labels. These statistics have been propagated to the info string of AdeftDisambiguator.
• Timestamps and additional metadata are collected at model training time, making changes in models more transparent. AdeftDisambiguator now has an additional method version based on some of this metadata.
• The ".adeft" folder containing models now has the version appended to it. E.g. in this release the folder will be named ".adeft_0.6.0". This will allow different versions of adeft with incompatible models to coexist on the same machine
• The command python -m adeft.download no longer takes argument --update, the new behavior is for all existing models to be replaced and a fresh copy of all existing models on S3 downloaded when the command is run.
• The method feature_importances of AdeftClassifier no longer raises an exception if called for a classifier trained before the information necessary to calculate feature importances was included. Now a warning is logged and None is returned