SHAP explainability

Explainable AI in Python

Fouad Trad

Machine Learning Engineer

Introduction to SHAP

SHAP → SHapley Additive exPlanations
Model-agnostic technique
Uses shapely values from game theory
SHAP values → quantify feature contributions to predictions

SHAP logo

Splitting profit among band members

Image for a musical band.

SHAP explainers

Compute SHAP values
Tree explainers → tree-based models

Image showing that SHAP explainers are divided between general explainers that can be applied to any model and type-specific explainers which are optimized for specific model types.

Admissions dataset

GRE Score	TOEFL Score	University Rating	SOP	LOR	CGPA	Chance of Admit	Accept
337	118	4	4.5	4.5	9.65	0.92	1
324	107	4	4	4.5	8.87	0.76	1
316	104	3	3	3.5	8	0.72	1
322	110	3	3.5	2.5	8.67	0.8	1
314	103	2	2	3	8.21	0.45	0

rf_reg: predicts chance of admit
rf_class: predicts acceptance
X: dataframe containing features

Deriving SHAP values

import shap

Regression

explainer_reg = shap.TreeExplainer(rf_reg)

shap_values_reg = explainer_reg.shap_values(X)

Classification

explainer_class = shap.TreeExplainer(rf_class)

shap_values_class = explainer_class.shap_values(X)

Understanding SHAP output

Regression

print(shap_values_reg.shape)

(400, 6)

Image showing SHAP values as a table having the same dimensions of the X dataframe (number of samples, number of features)

Understanding SHAP output

Regression

print(shap_values_reg.shape)

(400, 6)

Image showing SHAP values as a table having the same dimensions of the X dataframe (number of samples, number of features)

Classification

print(shap_values_class.shape)

(400, 6, 2)

Select values of positive class

positive_values = shap_values_class[:,:,1]

Feature importance

mean_shap_values_class = np.abs(shap_values_class[:, :, 1]).mean(axis=0)
mean_shap_values_reg = np.abs(shap_values_reg).mean(axis=0)

plt.bar(X_train.columns, mean_shap_values_class)
plt.bar(X_train.columns, mean_shap_values_reg)

Image showing feature importances for regression and classification, highlighting that CGPA and GRE scores are the most influential factors in predicting admissions.

Let's practice!

Explainable AI in Python