TfidfTransformer

Algorithm overview

TF-IDF reduces the influence of terms that occur in many documents while preserving terms that distinguish individual documents. TfidfTransformer is useful when counts already come from another source.

JavaScript implementation

@kanaries/ml accepts either number[][] or CSRMatrix counts and returns a sparse CSR result. This keeps text preprocessing memory-efficient in browser and Node.js pipelines.

Quick start example

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

const tfidf = new FeatureExtraction.TfidfTransformer({ norm: 'l2' });
const weighted = tfidf.fitTransform([[2, 0, 1], [0, 1, 1]]);
console.log(weighted.toDense());

Detailed API reference

new FeatureExtraction.TfidfTransformer({
  norm?: 'l1' | 'l2' | null; // 'l2'
  useIdf?: boolean;          // true
  smoothIdf?: boolean;       // true
  sublinearTf?: boolean;     // false
})

• fit(X): void learns inverse-document-frequency weights.
• transform(X): CSRMatrix applies term-frequency scaling, IDF, and normalization.
• fitTransform(X): CSRMatrix combines both operations.
• idf: number[] returns a defensive copy of the learned weights.

Inputs must contain non-negative term counts and use the fitted feature count.