Latent Dirichlet Allocation

Algorithm overview

LDA models each document as a mixture of latent topics and each topic as a distribution over words. It is useful for exploratory topic discovery after count vectorization.

JavaScript implementation

LatentDirichletAllocation accepts dense or CSRMatrix non-negative counts and performs batch variational updates entirely in browser or Node.js.

Quick start

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

const documents = ['cats chase mice', 'dogs chase balls', 'cats and dogs'];
const counts = new FeatureExtraction.CountVectorizer().fitTransform(documents);
const topics = new Decomposition.LatentDirichletAllocation({ nComponents: 5, randomState: 0 }).fitTransform(counts);
console.log(topics);

Detailed API reference

Options include topic priors, maxIter, maxDocUpdateIter, and meanChangeTol. components contains unnormalized topic-word parameters.