[shap-summary] SHAP Summary Plot

[MarkusNeusinger]

A SHAP (SHapley Additive exPlanations) summary plot showing feature impact on model predictions. Displays the distribution of SHAP values for each feature, ordered by importance, with color indicating feature value.

[claude]

✅ Spec ID: shap-summary

Rationale: SHAP summary plots are a distinctive visualization type in machine learning interpretability that show the distribution of SHAP values across features. This is different from bar-feature-importance which shows only aggregated importance values without the SHAP value distributions or feature value coloring.

Add label approved to begin generation.

[github-actions]

✅ Specification Ready: shap-summary

PR: #2934

---

specification.md

shap-summary: SHAP Summary Plot

Description

A SHAP (SHapley Additive exPlanations) summary plot displaying the distribution of SHAP values for each feature, ordered by mean absolute SHAP value (importance). Each dot represents a sample, positioned horizontally by its SHAP value and colored by the feature's value (typically low=blue to high=red). This visualization is essential for machine learning interpretability, showing both feature importance and the direction and magnitude of feature effects on model predictions.

Applications

• Explaining gradient boosting or random forest model predictions to stakeholders by showing which features drive predictions up or down
• Identifying non-linear relationships where high and low feature values both push predictions in the same direction
• Detecting feature interactions by observing clustered or bimodal SHAP value distributions

Data

• shap_values (numeric matrix) - SHAP values for each sample and feature, shape (n_samples, n_features)
• feature_values (numeric matrix) - Original feature values for coloring, same shape as shap_values
• feature_names (categorical) - Names of features for y-axis labels
• Size: 100-1000 samples recommended for clear distributions
• Example: SHAP values from shap.TreeExplainer on XGBoost/LightGBM model

Notes

• Sort features by mean absolute SHAP value (most important at top)
• Use diverging color scale (blue-red) to indicate low-to-high feature values
• Add vertical line at x=0 to clearly separate positive and negative impacts
• Consider jittering points vertically to reduce overlap
• For many features, show only top 10-20 most important

---

Next: Add approved label to merge this specification to main.

---

🤖 spec-create workflow

[github-actions]

🎨 matplotlib Preview

PR: #2980

---

🤖 impl-generate