RandomForestRegressor

Algorithm overview

RandomForestRegressor averages predictions from many decision trees. It is useful for nonlinear numeric prediction when a single regression tree is too noisy.

JavaScript implementation

@kanaries/ml exposes Ensemble.RandomForestRegressor for JavaScript and TypeScript projects. It supports bootstrap sampling, feature subsampling, and seeded randomness.

Interactive random forest regression playground

Change the number of trees, dataset, and noise below. The prediction line is produced by a live Ensemble.RandomForestRegressor; click the chart to add another training point.

Quick start example

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

const reg = new Ensemble.RandomForestRegressor({
  nEstimators: 25,
  maxDepth: 4,
  randomState: 7,
});

reg.fit([[0], [1], [2], [3]], [0, 1, 4, 9]);
const pred = reg.predict([[4]]);
console.log(pred);

Detailed API reference

new Ensemble.RandomForestRegressor(props?: {
  nEstimators?: number;
  bootstrap?: boolean;
  maxDepth?: number;
  minSamplesSplit?: number;
  maxFeatures?: number | 'sqrt' | 'log2';
  randomState?: number;
})

Methods:

• fit(trainX: number[][], trainY: number[]): void
• predict(testX: number[][]): number[]

Defaults are nEstimators: 100, bootstrap: true, and maxFeatures: 1.0.