fix: select Shapley Value column before aggregating in plot_global_feat

[daniel-n-silva]

Summary

plot_global_feat() raises TypeError: agg function failed [how->mean,dtype->object] under pandas >= 2.0.

Root cause

feat_data.groupby('Feature').mean()['Shapley Value'] aggregates every column first and selects Shapley Value afterwards. Global feature explanations produced by feat_explain_all always carry an Entity column, which ends up as object dtype whenever the entity identifier is a string (as in the AReM.ipynb example notebook) — feat_data.astype(...) never casts it. filter_dataset forwards this column unchanged into plot_global_feat. Older pandas silently dropped non-numeric columns before averaging; pandas >= 2.0 raises instead.

Fix

Select the Shapley Value column before aggregating, so the other columns are never touched:

avg_df = feat_data.groupby('Feature')['Shapley Value'].mean()

Identical result when no extra columns are present; works on both old and new pandas without needing numeric_only=True.

Fixes #62.

Test plan

• Reproduced the TypeError locally with pandas 2.3.3 using the real column set produced by feat_explain_all (Random Seed, NSamples, Feature, Shapley Value, Entity, Tolerance) with a string Entity column
• Confirmed the fix resolves it and returns the same values/Series.name/index.name as before
• Checked the other groupby(...).mean() call sites in src/ (event_level.py:104, pruning.py:92) — both already select/aggregate safely, no change needed there