node_neural_net

A utility to create and train a feed-forward neural network in Node.js

Examples

MNIST Handwritten Digit Recognition

let NeuralNet = require('node_neural_net');
let mnist = require('mnist');

// use a third party library to obtain training data
let set = mnist.set(8000, 2000);
let trainingSet = set.training;

// create input and output sets
let sample_in = new Array(8000);
let sample_out = new Array(8000);

for (let i = 0; i < 8000; ++i) {
    sample_in[i] = trainingSet[i].input;
    sample_out[i] = trainingSet[i].output;
}

// create a neural network with 784 input neurons (1 for each pixel),
//   2 hidden layers with 16 neurons each,
//   and 10 output neurons (1 for each digit).
let nn = new NeuralNet([784, 16, 16, 10]);

// nn.train(input_set, output_set, batch_size, learning_rate)
nn.train(sample_in, sample_out, 100, 0.1);

// save model information in a file called 'model.js' for later use
nn.save_weights('model.js');