- added
sklearn_crfsuite.metrics.flat_recall_score
.
- Properly close file descriptor in
FileResource.cleanup
; - declare Python 3.6 support, stop testing on Python 3.3.
- Small formatting fixes.
- scikit-learn dependency is now optional for sklearn_crfsuite; it is required only when you use metrics and scorers;
- added
metrics.flat_precision_score
.
- Ignore more errors in
FileResource.__del__
.
- Ignore errors in
FileResource.__del__
.
- Added
sklearn_crfsuite.metrics.sequence_accuracy_score()
function and relatedsklearn_crfsuite.scorers.sequence_accuracy
; FileResource.__del__
method made more robust.
backwards-incompatible:
crf.tagger
attribute is renamed tocrf.tagger_
; when model is not trained accessing this attribute no longer raises an exception, its value is set to None instead.new CRF attributes available after training:
classes_
size_
num_attributes_
attributes_
state_features_
transition_features_
Tutorial is added.
Initial release.