Changelog#

0.3.3 (2026-04-10)#

  • Added statistical stopping test for recall-based active learning screening

0.3.2 (2026-04-07)#

  • Added stratified split with fallback for rare classes in active learning

0.3.1 (2026-04-03)#

  • Multi-label support for cross-validation, metrics, and active learning

  • Best fold selection by validation loss; fixed formatting and edge cases

0.3.0 (2026-04-03)#

  • Human-in-the-loop description workflow: generate_descriptions(), confirm_descriptions()

  • Full-text semantic scoring for all records: score_full_text(), compare_screening_configs()

0.2.8 (2026-03-31)#

  • Suppressed safetensors load report during model initialization

0.2.7 (2026-03-19)#

  • Updated version references across all files

0.2.6 (2026-03-19)#

  • Fixed 0.0 returns where NaN is correct (screening, stopping, metrics, active learning)

  • Added bounds check in WSS calculation and cross-validation

  • Single-criterion fallback for stratified sampling

  • Narrowed exception handling in sample and screening modules

0.2.5 (2026-03-19)#

  • Multi-label iterative stratification for seed sampling

  • Improved active learning progress plot

0.2.3 (2026-03-19)#

  • Standardized variable names and print messages

  • Added active learning progress plot and persistent data splits

  • Added changelog page

0.2.2 (2026-03-18)#

  • Device fallback and warning fixes

  • Bounds validation in active learning simulation

0.2.1 (2026-03-17)#

  • Pattern match highlighting and conditional formatting for Excel

0.2.0 (2026-03-15)#

  • Multi-label classification, active learning, SAFE stopping, audit trails

  • Calibration, cross-validation, regex screening, stratified sampling

0.1.0 (2026-03-01)#

  • Initial release with PubMLP model and 6 embedding backends