SHAPExplainer<T>

Models & Types · AiDotNet.Interpretability.Explainers

Model-agnostic SHAP (SHapley Additive exPlanations) explainer using Kernel SHAP algorithm.

Implements: IGPUAcceleratedExplainer<T>, IGlobalExplainer<T, GlobalSHAPExplanation<T>>, ILocalExplainer<T, SHAPExplanation<T>>

For Beginners

SHAP values come from game theory and answer the question: “How much did each feature contribute to this specific prediction?”

Imagine you’re splitting a restaurant bill fairly among friends based on what each person ordered. SHAP does something similar - it fairly distributes the “credit” for a prediction among all input features.

Key properties of SHAP values:

  • They sum up to the difference between the prediction and the average prediction
  • Positive values mean the feature pushed the prediction higher
  • Negative values mean the feature pushed the prediction lower
  • The magnitude shows how important that feature was

This implementation uses Kernel SHAP, which works with ANY model by treating it as a black box.

Constructors

Constructor Summary
SHAPExplainer(Func<Matrix<T>, Vector<T>> predictFunction, Matrix<T> backgroundData, int nSamples, string[] featureNames, int? randomState) Initializes a new SHAP explainer with a prediction function and background data.

Properties

Property Summary
T BaselineValue { get; } Gets the baseline (expected) prediction value computed from background data.
bool IsGPUAccelerated { get; }
bool SupportsGlobalExplanations { get; }
bool SupportsLocalExplanations { get; }
int NumFeatures { get; } Gets the number of features being explained.
string MethodName { get; }

Methods

Method Summary
GlobalSHAPExplanation<T> ExplainGlobal(Matrix<T> data)
SHAPExplainer<T> FromModel(IFullModel<T, TInput, TOutput> model, Matrix<T> backgroundData, int nSamples, string[] featureNames, int? randomState) Initializes a new SHAP explainer with a model that implements prediction.
SHAPExplanation<T> Explain(Vector<T> instance)
SHAPExplanation<T>[] ExplainBatch(Matrix<T> instances)
void SetGPUHelper(GPUExplainerHelper<T> helper)