Shapash Files

What's Changed

New features

• Completely redesigned Smart Report system, with a new modular block-based architecture, default classification and regression reports, configurable report layouts, YAML configuration export, improved responsive rendering, new report components and better multiclass explainability/navigation. [#750] [#793] by @dalestee @guerinclement and @guillaume-vignal

• New contribution-weighted correlation plot to analyze relationships between feature contributions rather than only correlations between raw input features. Rows where one of the features has a strong contribution receive more weight in the correlation computation. [#754] by @guillaume-vignal

• Major improvements to interaction plots, including multiclass support, improved sampling and display ordering, adaptive color scales, density/volumetry information, better handling of categorical-like numerical variables and closer visual consistency with contribution_plot. [#758] by @guillaume-vignal

• Add waterfall visualization to local_plot, including baseline / SHAP base_values support, configurable x-axis behavior, improved tooltips and compatibility with LIME explanations. [#760] by @guillaume-vignal

• Add Lift Curve visualization to model evaluation plots and integrate it with Shapash styling. [#736] by @guillaume-vignal

SmartPredictor and production usage

• Strengthen SmartPredictor for production use: improve input validation and support pandas extension/nullable dtypes; add a save/load manifest containing Shapash, Python, model-framework and schema information; add structured prediction logging; and introduce lightweight schema-drift detection with configurable warnings. [#711] [#722] [#742] [#756] by @immu4989

Explainer architecture

• Refactor SmartExplainer into a lighter public facade, moving computation and state management into the dedicated Explainer class and reducing duplicated logic. Follow-up fixes remove redundant helpers and correct dependencies introduced by the refactor. [#746] [#774] [#789] by @guillaume-vignal and @guerinclement

Package and dependency support

• Simplify report dependencies following the Smart Report refactor, removing Jinja2, jupyter-client, nbconvert, notebook, papermill and pyarrow; the new report implementation now relies on panel. [#750] by @guerinclement

• Improve pandas 3 compatibility, particularly around the new string dtype behavior, and update the corresponding documentation. [#713] [#716] by @guerinclement and @guillaume-vignal

• Upgrade the WebApp to Dash >= 3.0.0. [#710] by @milton-minervino

• Update dependency compatibility, including category_encoders 2.10+, newer scikit-learn versions and removal of the Plotly upper bound; fix rendering regressions introduced by newer Plotly versions. [#740] [#748] [#770] [#771] by @meubleancien and @milton-minervino

• Temporarily cap pandas to avoid an upstream segmentation-fault issue. [#732] by @milton-minervino

• Update package license metadata to the SPDX Apache-2.0 format. [#783] by @guillaume-vignal

• Derive shapash.__version__ from package metadata instead of maintaining a hard-coded version. [#799] by @milton-minervino

Bug fixes

• Fix WebApp filtering with missing values. [#719] by @milton-minervino

• Fix threshold slider and clustering display issues in the WebApp. [#723] by @milton-minervino

• Improve confusion matrices, with a color scale better suited to imbalanced datasets and correct handling of integer class labels as discrete classes. [#738] [#739] by @ZakariaRida96

• Fix LIME multiclass contribution computation and subset handling. [#726] by @guillaume-vignal

• Prevent Explainer.filter() state from leaking into subsequent plots or to_pandas() calls. [#764] by @milton-minervino

• Honour the force parameter in compute_features_import. [#762] by @VenishPaneliya

• Make run_app().kill() correctly stop the WebApp server. [#737] by @milton-minervino

• Keep the regression True-vs-Predicted reference line on the actual y = x diagonal after zoom or axis reset. [#778] by @milton-minervino

• Improve stability explanations, fixing nearest-neighbor computation on small datasets and preserving neighbor ordering in explanations. [#791] [#794] by @aminemanai2003

• Add generic fallback fonts to plot font families for more reliable rendering across systems. [#780] by @milton-minervino

Testing, typing and code quality

• Enable and strengthen mypy type checking, including CI and pre-commit integration, fixes for the VS Code hook, broader type annotations and correction of previously undetected typing errors. [#744] [#751] [#787] [#796] [#798] by @milton-minervino and @guillaume-vignal

• Expand regression test coverage across explainers, plots and the WebApp. [#786] by @guillaume-vignal

• Revise the CI pipeline and pre-commit configuration. [#718] [#731] by @milton-minervino

• Add Codecov configuration, branch-coverage reporting and coverage badge. [#768] [#784] by @milton-minervino

• Fix Ruff issues and improve code formatting consistency. [#715] by @milton-minervino

• Remove duplicated and obsolete tests and clean test-generated files accidentally tracked by Git. [#741] [#766] by @milton-minervino and @Anai-Guo

Documentation and examples

• Add new example notebooks and refresh notebook tutorials. [#709] [#781] by @guillaume-vignal and @milton-minervino

• Move WebApp launch examples from documentation/notebooks into reusable scripts. [#779] by @milton-minervino

• Update and clarify the contribution guide. [#728] [#769] by @milton-minervino and @guerinclement

• Fix incorrect documented defaults and docstrings referring to non-existing parameters. [#765] [#767] by @VenishPaneliya

• Fix report documentation build and README link to the report notebook. [#800] by @milton-minervino

• Fix a typo in the Shapash architecture diagram. [#712] by @guerinclement

• Add P16 and Inria acknowledgements/logos. [#797] by @guillaume-vignal

Maintenance

• Merge the latest master changes back into the development branch. [#763] by @guillaume-vignal

New Contributors

• @dalestee made their first contribution in [#750]
• @milton-minervino made their first contribution in [#710]
• @meubleancien made their first contribution in [#740]
• @VenishPaneliya made their first contribution in [#767]
• @Anai-Guo made their first contribution in [#766]
• @aminemanai2003 made their first contribution in [#791]

Full Changelog: https://github.com/MAIF/shapash/compare/v2.9.0...v2.10.0