pw-lda

Linear discriminant analysis in JavaScript

Installation

npm install pw-lda

Getting Started

Two-dimensions are used in the below example, but any number of dimensions may be used.

LDA support both binary and multiclass classification. For multiclass classification, a pairwise voting system is used to extend the binary classifier to multiclass.

const LDA = require('pw-lda');
 
// Example with 2 classes
 
let class1 = [
    [0, 0],
    [1, 2],
    [2, 2],
    [1.5, 0.5]
];
 
let class2 = [
    [8, 8],
    [9, 10],
    [7, 8],
    [9, 9]
];
 
let classifier = new LDA(class1, class2);
 
let unknownPoints = [
    [-1, 0],
    [1.5, 2],
    [3, 3],
    [5, 5],
    [7, 9],
    [10, 12]
];
 
let predictions = [];
 
for(let i = 0; i < unknownPoints.length; i++){
    predictions.push(classifier.classify(unknownPoints[i]));
}
 
console.log(predictions); // [ 0, 0, 0, 1, 1, 1 ]
 
// Extending to a multiclass example
 
let class3 = [
    [-1, 10],
    [0, 12],
    [1, 11],
    [0.5, 9]
];
 
unknownPoints = unknownPoints.concat([
    [0, 11],
    [-1, 8],
    [1, 9]
]);
 
classifier = new LDA(class1, class2, class3);
 
predictions = [];
 
for(let i = 0; i < unknownPoints.length; i++){
    predictions.push(classifier.classify(unknownPoints[i]));
}
 
console.log(predictions); // [ 0, 0, 0, 1, 1, 1, 2, 2, 2 ]

Documentation

Documentation is available at http://pwstegman.me/pw-lda/