Add shapiq -> shap.Explanation bridge helper

[adrian-prior]

Summary

Adds a small bridge helper so users who want to compute Shapley values with shapiq (faster and TabPFN-friendly) but plot with the shap library (mature plotting ecosystem) don't have to write the 15-line "loop / stack / average baseline / construct Explanation" boilerplate themselves.

What changed

New module — src/tabpfn_extensions/interpretability/shap.py:

def shapiq_to_shap_explanation(
    explainer, X, *, budget, feature_names=None,
) -> shap.Explanation:
    ...

Mirrors the pattern in examples/interpretability/shap_example.py: one .explain(...) per row, stack the first-order arrays into (n, d), average baseline values, return a shap.Explanation. The import shap is inside the function body so the rest of the interpretability surface stays usable without shap installed.

That filename was freed up when #283 removed the legacy SHAP-based explainer; reusing it for the bridge helper maps naturally (shap.py = shap library integration, shapiq.py = shapiq library integration).

Re-export — interpretability/__init__.py now exports shapiq_to_shap_explanation so the import is short:

from tabpfn_extensions.interpretability import shapiq_to_shap_explanation

Example shrunk — examples/interpretability/shap_example.py: the previous 15-line boilerplate collapses to a single shapiq_to_shap_explanation(...) call. The five shap.plots.* demonstrations are unchanged.

README — src/tabpfn_extensions/interpretability/README.md: the paragraph that mentioned "or convert the values to shap.Explanation and use shap.plots.*" now names the helper explicitly and shows the one-liner, with the "use shapiq's native plots" path still presented as the alternative.

What this is not

• Not adding shap to the interpretability extra. Same decision [RES-1467] Drop SHAP, refresh shapiq examples around v3 KV cache #283 made — shapiq is the runtime dep, shap is opt-in for plotting. Documented in the helper's docstring.
• Not supporting higher-order interactions. shap.Explanation doesn't represent them; for those, users should keep using shapiq's native plots. Documented in the docstring.

Verification

Smoke-tested end-to-end in a clean venv (Python 3.12, shap + shapiq + tabpfn-extensions installed from local source) with a stub explainer that returns shapiq.InteractionValues. Checks:

Property	Result
shapiq_to_shap_explanation(...) import path	✅
Returned object is shap.Explanation	✅
values.shape == (n, d)	✅
base_values.shape == (n,)	✅
feature_names passed through	✅

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces a bridge helper function, shapiq_to_shap_explanation, to facilitate the use of the SHAP library's plotting ecosystem with Shapley values computed via shapiq. The changes include the implementation of the helper in a new shap.py module, updates to the interpretability documentation, and a refactor of the SHAP example script to utilize this new utility. Feedback was provided to improve the robustness of the helper by handling empty input cases and using per-row baseline values instead of a global average to better support diverse explainer configurations.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.