ClassifierChain

Algorithm overview

A classifier chain fits one binary classifier per output. Each member sees the original features plus earlier labels, allowing later predictions to depend on earlier outputs rather than assuming label independence.

JavaScript implementation

Pass any registered single-output classifier to MultiOutput.ClassifierChain. A fixed, identity, or seeded random order is supported, and optional cross-validated training extensions reduce target leakage.

Quick start example

import { Bayes, MultiOutput } from '@kanaries/ml';

const X = [[0], [1], [2]];
const testX = [[1.5]];
const chain = new MultiOutput.ClassifierChain({
  estimator: new Bayes.GaussianNB(), order: [0, 1],
});
chain.fit(X, [[0, 0], [0, 1], [1, 1]]);
const labels = chain.predict(testX);
console.log(labels);

Detailed API reference

Options: estimator, order?: number[] | 'random', cv?: number | null, and randomState. Methods: fit(X, Y), predict(X), predictProba(X), and exact-match score(X, Y). chainOrder and estimators return defensive arrays. predictProba requires the base classifier to expose probabilities and returns the positive-class probability per output.