ComplementNB

Algorithm overview

ComplementNB is a naive Bayes variant designed for non-negative count features and often useful on imbalanced text-like classification problems. It estimates feature weights from the complement of each class.

JavaScript implementation

@kanaries/ml exposes Bayes.ComplementNB so JavaScript and TypeScript applications can run this lightweight classifier in browser or Node.js environments.

Quick start example

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

const X = [[3, 0, 1], [2, 0, 1], [0, 2, 2], [0, 3, 1]];
const y = [0, 0, 1, 1];

const clf = new Bayes.ComplementNB({ alpha: 1.0, norm: false });
clf.fit(X, y);
const pred = clf.predict([[2, 0, 1], [0, 2, 1]]);

Detailed API reference

new Bayes.ComplementNB(props?: {
  alpha?: number;
  forceAlpha?: boolean;
  fitPrior?: boolean;
  classPrior?: number[] | null;
  norm?: boolean;
})

Methods:

• fit(X: number[][], y: number[]): void
• predict(X: number[][]): number[]

Feature values must be non-negative.