v0.20.2

What's Changed

• Allow using predefined configurations when installing from PyPi

v0.20.1

What's Changed

• Add missing configuration file

v0.20.0

What's Changed

• Add SBERT based multiclass SVM model. Set it to the default for watsonx data collection users by @alonh in #632
• Mutliclass iteration triggering logic improvement by @martinscooper in #630
• Multiclass category menu UX and logic improvement by @martinscooper in #631

Full Changelog: v0.19.0...v0.20.0

v0.19.0

New features

• Improve multiclass UX design by @martinscooper and @natti in #605
• Add SBERT representation and the SVM over SBERT model. Refactor memory and cache mechanism to support vectors by @alonh in #626
• Add multiclass ensemble model by @arielge in #595
• Update python version and dependencies by @arielge in #624
• Add tutorial for multiclass workspaces by @martinscooper in #592, #593 and #594.
• Adapt force_train for multiclass workspaces by @martinscooper in #600
• Add percentage to panel filter dropdowns by @martinscooper in #619
• Split datasets with no document id into 10k parts by @alonh in #622

Improvements

• UI text overflow improvements by @martinscooper in #601
• Adapt labeling status message and model checks conditions by @martinscooper in #613

Fixes

• Dataset delete checkbox remains checked after submit by @martinscooper in #591
• Map multiclass model scores to the corresponding class ids by @arielge in #606
• Do not allow changing category name to a name that already exists by @arielge in #599
• User can't edit category description or color without changing the category name by @martinscooper in #603
• User can bypass validation using shortcuts by @martinscooper in #610
• Don't select 'sort' text when double clicking on the sorting arrow by @martinscooper in #612
• Deleting workspaces fails if workspace isn't in memory by @martinscooper in #614
• Status call logic is wrong by @martinscooper in #615
• Create new iteration if there is an error in the should_train condition and fix test race condition by @alonh in #617
• Fix force train test by @martinscooper in #620
• Bugfix in accessing sorted_class_by_freq when there's no training data by @lenadankin in #616
• Escape with slash only as a fallback on import data. Convert input to string to handle float values by @alonh in #621
• Percentage string is NaN is there are no labeled elements by @martinscooper in #625
• Training logic doesn't trigger new model iterations when appropriate by @martinscooper in #627 and #628

Chore

• Change create workspace parameter name by @martinscooper in #596
• Rename /set_element_label endpoint parameters and refactor by @martinscooper in #618

Full Changelog: v0.18.0...v0.19.0

v0.18.0

What's Changed

New features

• Multiclass mode (beta) implementation
• User labels and model predictions panel filters

Improvements

• Improve left sidebar panels responsiveness

Full Changelog: v0.17.2...v0.18.0

v0.17.2

What's Changed

• fix(backend): user labels are not presented on doc view

v0.17.1

What's Changed

• feat: change to search panel when iteration status is INSUFFICIENT_TAIN_DATA by @martinscooper in #584

Full Changelog: v0.17.0...v0.17.1

v0.17.0

What's Changed

• feat(backend): allow export and import of label metadata and log errors by @alonh in #583

Full Changelog: v0.16.1...v0.17.0

v0.16.1

What's Changed

• Add insufficient train data messages customization by @martinscooper in #582

v0.16.0

What's Changed

• feat(backend): add INSUFFCIENT_TRAIN_DATA status and save iteration and label source information by @alonh in #579
• New sidebar elements and positive prediction design by @martinscooper in #577
• Handle new status iteration and send more information on labeling endpoint by @martinscooper in #580
• feat(backend): update insufficient train data text by @alonh in #581

Full Changelog: v0.15.1...v0.15.2