Multilayer Perceptron. Simple neural network. Written in ES6 (Javascript).

Usage

1. Create new Network and Trainer objects.

let network = new Network(2, 3, 1),
    trainer = new Trainer()

Network constructor accepts amounts of input, hidden and output neurons.

2. Train network

let weights = trainer.train(network, patterns[, learnRate])

train method returns weights. Save it to the file as json then next time you can use it simply passing to setWeights method of Network class. patterns is an array of patterns, contains input and desired values for each inputs and outputs:

[
  {
    "input": [ 0, 0 ],
    "desired": [ 0 ]
  }
]

3. Use

network.calculate(input)

input object must be the same as input property of the pattern used in training

P.S.

See train.js as an example of training. Run node train.js and it will save calculated weights to the file. Then run node process.js and it will execute neural network using calculated weights.