neur

Simple neural network implementation in JS

Installation

Install via npm:

npm install neur

Usage

Step 1: Initialize a network

var Neural = require('neur');
var neural = Neural();

or

var neural = require('neur')();

You can config the network by passing the options to the constructor:

var neural = Neural({
  learningRate: 0.7,
  iterations: 10000,
  hiddenUnits: 3
});

By default, we have a network with 1 hidden layer which has 3 neurons in total. And it will train the input data 10,000 times.

Step 2: Train a network

neural.learn([
  { input: [ <input-values> ], output: [ <output-values> ] },
  ...
]);

Step 3: Predict

var result = neural.predict([ <predict-input-values> ]);

Examples

Example 1: Basic using

var Neural = require('neur');
 
var neural = Neural()
    .learn([
      { input: [0, 0], output: [0] },
      { input: [0, 1], output: [1] },
      { input: [1, 1], output: [1] },
      { input: [1, 0], output: [0] }
  ]);
 
console.log(neural.predict([1, 0])); // ~0
console.log(neural.predict([1, 1])); // ~1

Example 2: Use Model mapping to train complex data

var Neural = require('neur');
 
var color = Neural().model({ r: 0, g: 0, b: 0 });
var guess = Neural().model({ black: 0, white: 0 });
 
var result = Neural()
    .learn([
      {
          input: color.in({ r: 0.03, g: 0.7, b: 0.5 }),
          output: guess.in({ black: 1, white: 0 })
      },
      {
          input: color.in({ r: 0.16, g: 0.09, b: 0.2 }),
          output: guess.in({ black: 0, white: 1 })
      },
      {
          input: color.in({ r: 0.5, g: 0.5, b: 1.0 }),
          output: guess.in({ black: 0, white: 1 })
      }
    ])
    .predict(color.in({ r: 1, g: 0.4, b: 0 }));
 
console.log(guess.out(result)); // { black: ~0, white: ~1 }