Simple neuron

neuron.js

/* activation functions */

// linear activation (no-op)
function linear(value) { return value }

// threshold activation
function threshold(x = 0) {
  return (value) => value >= x ? 1 : 0
}

// rectified/ReLU activation
function rectified(x = 0) {
  return (value) => Math.max(x, value)
}

/* utility functions */
function sum(list, mapFn) {
  return list.reduce((acc, x, i) => {
    return acc + mapFn(x, i)
  }, 0)
}

// define a neuron
//
// bias: integer
// weights: integer[n]
// activationFn: function
//
// returns: function(...integer[n])
function neuronFactory(bias, weights, activationFn = linear) {
  return (...inputs) => {
    // verify that parameter size is right
    if (weights.length != inputs.length) {
      throw new Error(`Invalid length. Expected ${weights.length} arguments, got ${inputs.length}.`) 
    }

    // sum the total of each input multiplied by weight
    const total = sum(inputs, (input, i) => input * weights[i])

    // compute the result value using the sum + bias
    return activationFn(total + bias)
  }
}

// define a neuron: bias, weights, activation function
let neuron = neuronFactory(3, [1, 1, 3], linear)
// use the neuron, and print result
console.log(neuron(1, 2, 3))

// define a neuron
neuron = neuronFactory(10, [1, 2, 3, 8], rectified(0))
// use the neuron, and print result
console.log(neuron(5, 2, 3, 3))