v0.4.0

Changes

• Add select_consensus_statements() function, and wire into Polis implementation.
• Allow calculate_comment_statistics() to work without groups/labels.
• Generalize format_comment_stats() to work for group and consensus statements.
• Add select_representative_statements() to PolisClusteringResult as repness key.
• Rename arg pick_n to pick_max in select_consensus_statements(), for clarity and consistency.
• Slight change to PolisRepness type, so group IDs now returned as ints.
• Add print_selected_statements() presenter for inspecting PolisClusteringResult.
• Add print_consensus_statements() presenter for inspecting PolisClusteringResult.
• Allow pick_max and confidence interval args to be set in polis.run_clustering().
• Allow get_corrected_centroid_guesses() to unflip each axis if correction not needed.
• Abstracted reducer and clusterer algorithm support.
  • Added support for pacmap/localmap beyond PCA.
  • Added support for HDBSCAN clustering beyond KMeans.
  • Allow passing of arbitary params into reducer/clusterer.
• Remove support for polis_legacy implementation (PolisClient).
• Added disagree variant of group-informed-consensus. (group-informed-consensus-disagree)
• Brought group-informed-consensus metrics to top-level result object.
• Renamed run_clustering function to run_pipeline and created base pipeline implementation.
• Add option to generate_figure_polis to configure showing pid labels (show_pids).
• Remove deprecated methods from doc website.
• Remove deprecated modules from prior import paths.
• Avoid using dataframes in a few low level util function, in favour of numpy arrays.
• Rename projected_{participants,statements} to {participant,statement}_projections in run_pipeline results. Also coords keyed to ID, instead of dataframes.
• Remove agora implementation and tests. (#73)
• Migrate from reference HDBSCAN module (in scikit-learn) to full-featured HDBSCAN* package.
• Add dependency groups to avoid installing everything. (#11)
• Add support for statement and participant IDs to be strings. (#92)
• Added BestPolisKMeans estimator. (#111)

Fixes

• Handle when is-meta and is-seed columns arrive in CSV import.
  #55
• Handle loading comments data from API when is_meta missing in CSV import.
• Only pass unique labels into generate_figure() colorbar.
• bugfix: clusterer_kwargs and reducer_kwargs were not being pass through run_pipeline().
• bugfix: Ensure run_pipeline() passes random_state to reducer.
• bugfix: Fix overly constrained versions from #80.
• bugfix: Ensure we don't crash when a participant ID in keep_participant_ids doesn't exist in vote matrix. (#100)
• bugfix: Relaxed weight_x_32767 field to accept alternative responses (voxit server). (#107)

Chores

• Update the release process instructions.
• Added simulate_api_response() test helper for easier comparison with polismath output.
• Bumped minimum supported Python version from 3.8 to 3.10. (#94)
• Added CI testing for all of Python 3.10, 3.13. (#93)
• Enforced a build of website and notebooks in pull requests. (#96)
• Added support for test fixtures that pull data from live polis convos. (#98)
• Added data loader to docs website. (#101)
• Sync'd sanity-check API tests with new Pol.is server response fields. (#106)
• Resolved deprecated method in HDBSCAN. (#109)
• Removed CloudflareBypassHTTPAdapter that is no longer necessary. (#108)
• Extracted Exporter class out of data_loader into data_exporter. (#100)
• Updated HDBSCAN to allow using scikit-learn >=1.8. (#112)
• Keep running CI even when one version of Python fails. (#112)