AREkit-0.20.4
·
53 commits
to 0.20.4-rc
since this release
Release Notes
- Labels conversion
to_str
andfrom_str
now a part of external formatters (unique for each source, experiment, etc.). - Added labels-scaler, and labels casing (to int or uint) now depends on scaler;
- Added bert exporter in contribution folder: with related formatters according to the [paper]:
- NLI -- (Natural language inference) format, assumes to provide an additional sentence, which describes
attitude should be extracted - QA -- (Question answering) provides an additional question onto attitude sentiment.
With Label encoding in following format: - Multiple -- all the supported sentiment labels (positive, negative, neutral)
- Binary -- (YES, NO) according to mention (additional sentence), provided by NLI and QA formatters.
- NLI -- (Natural language inference) format, assumes to provide an additional sentence, which describes
- Refactoring experiments in order to apply the latter also for classifiers (models from scikit-learn)
- Updated nn-engine API
- Refactoring tf-based neural network implementation.
- Bert now moved into separated folder from
contrib
directory. - frame_variants moved to
frames
directory. - Frame variants labeling in news now performed during
parse
operation. DataType
now enumeration. List of Supported data-types now a part of experiment
The latter were moved onto sample level.Service
folder removed as the latter assumes to be apart of this repository.