PrototypeExplainer<T>

Models & Types · AiDotNet.Interpretability.Explainers

Prototype-based explainer that explains predictions using similar examples.

Implements: ILocalExplainer<T, PrototypeExplanation<T>>

For Beginners

Sometimes the best explanation is an example! This explainer answers “Why this prediction?” with “Because it’s similar to these examples.”

How it works:

  1. Maintain a set of “prototypes” (representative examples from training data)
  2. For a new prediction, find the most similar prototypes
  3. Explain the prediction based on these similar examples

Types of prototype explanations:

  • Nearest neighbors: Show the K closest training examples
  • Same-class prototypes: Show similar examples with the same prediction
  • Contrast prototypes: Show similar examples with different predictions

Why prototypes are useful:

  • Intuitive: “This loan was approved because it’s similar to John’s approved loan”
  • Concrete: Shows actual examples, not abstract feature weights
  • Trustworthy: Users can verify the similarity themselves

Example use cases:

  • Medical diagnosis: “This case is similar to these past cases that were diagnosed as…”
  • Credit decisions: “Your application is similar to these approved/denied applications”
  • Image classification: “This image looks like these training images of cats”

Constructors

Constructor Summary
PrototypeExplainer(Func<Matrix<T>, Vector<T>> predictFunction, Matrix<T> prototypes, Vector<T> prototypeLabels, int numNeighbors, DistanceMetric distanceMetric, string[] featureNames, string[] prototypeNames) Initializes a new Prototype explainer.

Properties

Property Summary
bool SupportsGlobalExplanations { get; }
bool SupportsLocalExplanations { get; }
string MethodName { get; }

Methods

Method Summary
PrototypeExplanation<T> Explain(Vector<T> instance) Explains a prediction using similar prototypes.
PrototypeExplanation<T>[] ExplainBatch(Matrix<T> instances)