forwardjs-ml-math-lodash.js

const map = require('lodash/map')
const reduce = require('lodash/reduce')

const validateLengths = (a,b,action) => {
  if (a.length !== b.length) {
    throw Error(`Can\'t ${action} arrays of length ${a.length} and ${b.length}`);
  }
}

const arrayMultiply = (arr1, arr2) => {
  validateLengths(arr1, arr2, "multiply")
  return reduce(arr1, (accumulator, arrayItem, i) =>
    accumulator += arrayItem * arr2[i]
  , 0)
}

const arrayAdd = (arr1, arr2) => {
  validateLengths(arr1, arr2, "add")
  return map(arr1, (arrayItem, i) =>
    arrayItem + arr2[i]
  )
}

const arraySubtract = (arr1, arr2) => {
  validateLengths(arr1, arr2, "subtract")
  return map(arr1, (arrayItem, i) =>
    arrayItem - arr2[i]
  )
}

const sigmoid = (z) =>
  1 / (1 + Math.exp(z * -1))

const sigmoidGradient = (z) =>
  sigmoid(z) * (1 - sigmoid(z))

const sigmoidInverse = (a) =>
  Math.log(a) - Math.log(1 - a)

exports.arrayMultiply = arrayMultiply;
exports.arrayAdd = arrayAdd;
exports.arraySubtract = arraySubtract;
exports.sigmoid = sigmoid;
exports.sigmoidGradient = sigmoidGradient;
exports.sigmoidInverse = sigmoidInverse;

PS. haven't tested for any regressions other than getting similar output from the network