Local explainability with SHAP

Explainable AI in Python

Fouad Trad

Machine Learning Engineer

Global vs. local explainability

Global explainability

Overall model behavior
Doesn't explain individual instances

Professor explaining the general performance of the classroom to all students.

Local explainability

Explains prediction for specific data point
Crucial for sensitive applications

Professor explaining to a specific student the reason behind their grade.

¹ Images generated by DALL-E

Heart disease dataset

age	sex	blood_pressure	ecg_results	thalassemia
52	1	125	1	3
53	1	140	0	3
70	1	145	1	3
61	1	148	1	3
62	0	138	1	2

knn: KNN classifier predicting risk of heart disease

Local explainability with SHAP

explainer = shap.KernelExplainer(knn.predict_proba, shap.kmeans(X, 10))


test_instance = X.iloc[0, :]


shap_values = explainer.shap_values(test_instance)


print(shap_values.shape)

(6, 2)

SHAP waterfall plots

Shows how features increase or decrease model's prediction

A waterfall plot that shows multiple features and the contribution of each towards the prediction.

SHAP waterfall plots

Shows how features increase or decrease model's prediction

Same waterfall plot showing the baseline value, which serves as an initial point for the prediction.

SHAP waterfall plots

Shows how features increase or decrease model's prediction

Same waterfall plot showing that in waterfall plots, features pushing to the right increase the prediction while features pushing to the left decrease it.

Creating waterfall plots

shap.waterfall_plot(

  shap.Explanation(

    values=shap_values[:,1],

    base_values=explainer.expected_value[1],

    data=test_instance,

    feature_names=X.columns

  )

)

Waterfalls for several instances

Figure displaying four waterfall plots, one for each sample, illustrating how different samples follow distinct patterns to reach the model’s decision.

Let's practice!

Explainable AI in Python