docs(notebook): Improve notebook for v3

[adrian-prior]

Summary

Updates to examples/notebooks/TabPFN_Demo_Local.ipynb:

• Shapley Values section (cells 34–36): switches from the legacy interpretability.shap adapter (classic shap library + PermutationExplainer) to shapiq via get_tabpfn_imputation_explainer. Same imputation paradigm, but using shapiq's native bar_plot, beeswarm_plot, and plot_force. Surrounding markdown rewritten to explain shapiq and link to shap_example.py in tabpfn-extensions for the classic-shap path.
• KV cache enabled on the SHAP demo classifier (fit_mode="fit_with_cache") since shapiq issues many predict_proba calls against the same training set. Markdown notes that this is local-only — for the client backend, drop the flag (slower).
• Switched the explainer to imputer="baseline" for a better speed/quality trade-off, kept n_estimators=8 with a comment about throughput on stronger GPUs, and right-sized the demo split to n_samples_test=5, n_samples_train=200.
• Feature selection cell (cell 40): added cv=3 to feature_selection(...) to make the example reproducible without depending on the library's default CV setting.
• Time-series install cell (cell 53): removed the raise ValueError(...) that interrupted the run after install; replaced with a clearer markdown note telling the reader to restart the runtime before continuing.
• Causal-inference install cell (cell 73): switched !pip install → !uv pip install for consistency with the rest of the notebook.
• Added a small Feedback cell at the end pointing readers to Discord and GitHub.

Test plan

• Run the updated Shapley Values cell end-to-end (Colab GPU or local CUDA) and confirm the bar, beeswarm, and force plots render.
• Run the feature selection cell and confirm it executes with cv=3.
• Re-run the time-series section: install cell completes, then after a runtime restart the rest of the section runs cleanly.
• Re-run the causal inference install via uv pip and confirm both tabpfn and econml are picked up.
• Quick top-to-bottom smoke run of the notebook to confirm no broken imports were introduced.

🤖 Generated with Claude Code

[gemini-code-assist]

Note

Gemini is unable to generate a review for this pull request due to the file types involved not being currently supported.

[LeoGrin]

Thanks!!!

[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.