ml4node

A basic machine learning node.js library written to learn javascript. The two main modules are Vector and Matrix, which implement the linear algebra necessary for building machine learning models.

Neural Net Example (ml4node/scripts/nn.js)

'use strict'; 

const Matrix = require('../lib/matrix');
const sigmoid = require('../lib/math').sigmoid;

  var ins = new Matrix([
    [0, 0, 1],
    [0, 1, 1],
    [1, 0, 1],
    [1, 1, 1]
  ]);

var out = new Matrix([[0, 1, 1, 0]]).transpose();

var trainNeuralNetwork = function(input, output, iters) {
var sd = 2;
var mean = -1;

var inputWeights = Matrix.create(input.cols, input.rows)(
   () => sd * Math.random() + mean
   ); 
var hiddenWeights = Matrix.create(output.rows, output.cols)(
   () => sd * Math.random() + mean
   );

var inputLayer = input.mMult(inputWeights).map(sigmoid(false));
var hiddenLayer = inputLayer.mMult(hiddenWeights).map(sigmoid(false));

for (var i = 0; i < iters; i++) {
   inputLayer = input.mMult(inputWeights).map(sigmoid(false));
   hiddenLayer = inputLayer.mMult(hiddenWeights).map(sigmoid(false));

    var hiddenError = output.subtract(hiddenLayer).multiply(
      hiddenLayer.map(sigmoid(true))
      );
    var inputError = hiddenError.mMult(hiddenWeights.transpose()).multiply(
      inputLayer.map(sigmoid(true))
      );

    hiddenWeights = hiddenWeights.add(
      inputLayer.transpose().mMult(hiddenError)
      );
    inputWeights = inputWeights.add(input.transpose().mMult(inputError));
  }
 return hiddenLayer;
};

var testResults = trainNeuralNetwork(ins, out, 50000);
console.log(testResults);

Output:

Matrix {
  elements:
   [ 0.002796177648568713,
     0.996543313711873,
     0.9964837863426119,
     0.004252828412771536 ],
  rows: 4,
  cols: 1 }

At this time I suggest you DO NOT use in production.

pgerbes1/ml4node

ml4node

Neural Net Example (ml4node/scripts/nn.js)