ConceptActivationVector<T>

Models & Types · AiDotNet.Interpretability.Explainers

Represents a Concept Activation Vector.

For Beginners

A CAV is a vector in the layer activation space that points in the direction of “more concept.” It’s the key component of TCAV that allows us to measure concept influence on predictions.

Constructors

Constructor Summary
ConceptActivationVector(Vector<T> weights, double classifierAccuracy) Initializes a new Concept Activation Vector.

Properties

Property Summary
Vector<T> Weights { get; } Gets the CAV weights (direction in activation space).
double ClassifierAccuracy { get; } Gets the accuracy of the linear classifier used to train this CAV.