SHAP

Shapley Additive exPlanations, Shapley Values

• The goal of SHAP is to explain the prediction of an instance x by computing the contribution of each feature to the prediction

• Build around idea of
  Shapley values (assume feature is a player and combination as colition)

pip install shap

import shap

explainer = shap.Explainer(classifier)
shap_values = explainer(counterfactuals)

shap_values_stereotype = shap_values[:, :, "stereotype"].values
shap_vectors = []

# Save SHAP values in vectors for subsequent calculation
for index, values in enumerate(shap_values_stereotype):
    # Trim to exclude whitespace and punctuation 
    trimmed_values = values[1:-2]
    shap_vectors.append(trimmed_values)
    print(f"Sentence {index+1} SHAP vector: {trimmed_values}")

shap.plots.text(shap_values[:, :, "stereotype"])

SHAP variants

• KernelSHAP

• TreeSHA

SHAP interpretation methods

• SHAP Feature Importance

• SHAP Summary Plot

• SHAP Dependence Plot

• SHAP Interaction Values

Shapley Values: Unlocking Intuition with Venn Diagrams

Shapley values are an extremely popular tool in explainable AI. For reference, check out the Python package SHAP.

https://medium.com/@carson.loughridge/shapley-values-unlocking-intuition-with-venn-diagrams-86e76d8c99c5

Welcome to the SHAP documentation — SHAP latest documentation

SHAP (SHapley Additive exPlanations) is a game theoretic approach to explain the output of any machine learning model. It connects optimal credit allocation with local explanations using the classic Shapley values from game theory and their related extensions (see papers for details and citations).

https://shap.readthedocs.io/en/latest/