Add Contributions Projection Plot to the Shapash HTML Report

[guillaume-vignal]

Description

The new contributions_projection_plot visualization projects each instance’s Shapley values into a 2D space, offering deep insight into why predictions are correct or incorrect.

Currently, the Shapash Report (generated via SmartExplainer.report) includes plots such as:

• True vs Predicted plots,
• Feature importance,
• Local explanations.

However, it does not yet include the Contributions Projection Plot, which could add significant value to static model analysis.

Feature Proposal

• Integrate the Contributions Projection Plot into the Shapash HTML report.

• Include three versions of the plot (or a toggle):

  1. Projection colored by prediction (or probability score).
  2. Projection colored by true values.
  3. Projection colored by prediction error.

• Allow users to visually compare these plots to identify clusters of well/poorly predicted instances.

Expected Benefits

• Provide a static yet rich view of the Shapley-based projections for stakeholders who consume the report without using the WebApp.
• Make it easy to identify zones of high error or unusual behavior without running live code.

Implementation Notes

• Reuse the contributions_projection_plot logic from SmartPlotter.
• Embed them into the HTML report.