// http://natureofcode.com

​

// Neural Network: Perceptron

​

// A Simple Perceptron Class

// Includes functionality for user to train network with known data,

// and generate an output based on an array of inputs

​

class Perceptron {

  constructor(num_of_inputs) {

    this.weights = [];

    this.c = 0.00002; // learning constant

​

​

    // Weights start off random 

    for (let i = 0; i < num_of_inputs; i++) {

      this.weights[i] = Math.random(-1, 1);

    }

  }

​

​

  // Return an output based on inputs

  feedforward(inputs) {

    let sum = 0;

    for (var i = 0; i < this.weights.length; i++) {

      sum += inputs[i] * this.weights[i];

      // console.log(i + ": " + inputs[i] + ", " + this.weights[i]);

    }

    // console.log("sum: " + sum);

    return this.activate(sum);

  }

​

​

  // Output is +1 or -1

  activate(sum) {

    if (sum > 0) return 1;

    else return -1;

  }

​

​

  // Train network against known data

  train(inputs, desired) {

    const guess = this.feedforward(inputs);

    const error = desired - guess;

​

    for (let i = 0; i < this.weights.length; i++) {

      this.weights[i] += this.c * error * inputs[i];

    }

​

  }

​

  // Return weights

  getWeights() {

    return this.weights;

  }

​

}