edubehaviors-kit 0.1.1

Bug fix and QOL release.

Fixed

• AssertionAnnotator with lazy=True no longer accumulates SetFit models on the GPU. Annotator now correctly collects each model after prediction.

Added

• edubehaviors.__version__ attribute exposes the installed package version.

Changed

• ClassificationPipeline raises a ValueError if the label column contains missing values.

Documentation

• New words page listing the default WordAnnotator word list.
• The assertions table now reports test set support for each assertion.
• The example notebook now runs straight on Colab.
• The example dataset (examples/talkmoves.csv, formerly talkmoves_tutor.csv) now includes student utterances alongside tutor utterances.

edubehaviors-kit 0.1.0

First release on PyPI.

Added

Annotators

• WordAnnotator: Annotates text data with either counts or true/false appearances of
  a list of words. Ships with a list of common words derived from classroom transcripts.
• AssertionAnnotator: Predicts assertions—short, specific behaviors or attributes—on
  text data using public encoder models. Includes 49 published assertion classifiers,
  described on the assertions page of the documentation.

Modeling

• standard_classifier: Gives a good baseline version of LogisticRegressionCV
  preconfigured sensible settings for education data classification.
• ClassificationPipeline: An easy-to-use class that annotates a labelled frame, splits
  it, fits a classifier and evaluates it all in one call.

Requirements

• Python 3.12, 3.13 or 3.14.
• scikit-learn is pinned to ==1.7.2, the version the published SetFit heads are pickled
  with. Newer releases warn on unpickling.

Feedback

This is an early release and the API may still change. Bug reports and feature requests are
welcome at https://github.com/scale-nssa/edubehaviors-kit/issues.