multipleFeaturesGradientDescent.js

/* ----------------------------------------- *
        Learning Machin Learning
* ----------------------------------------- */
// gradientDescent of the square error function for a Theta of any number of
// parameters


// multiplyVectors :: List number -> List number -> number
const multiplyVectors = (xs, ys) =>
  xs.reduce((acc, x, i) => acc + x * ys[i], 0)

// sum :: (a -> Int) -> List a -> Int
const sum = (transformer, list) =>
  list.reduce((acc, item) => acc + transformer(item), 0)


  // getNextTheta :: List Number -> List Data -> Theta -> Theta
const getNextTheta = (step, dataSet) => theta =>
  theta.map((currentTheta, i) =>
    currentTheta - step[i] / dataSet.length * sum(([xs, y]) => (multiplyVectors(xs, theta) - y) * xs[i], dataSet)
  )

// isConverging :: Theta -> Theta -> Bool
const isConverging = (xs, ys) =>
  xs.reduce((acc, x, i) => acc && Math.abs(x - ys[i]) < 0.001, true)


// repeatUntilConverge :: (Theta -> Theta) -> Theta -> Theta
const repeatUntilConverge = (f, theta) => {
  const nextTheta = f(theta)
  return isConverging(nextTheta, theta)
    ? nextTheta
    : repeatUntilConverge(f, nextTheta)
}

// gradientDescent :: { step :: List Number, dataSet :: List Data } -> Theta
const gradientDescent = ({ step, dataSet }) =>
  repeatUntilConverge(getNextTheta(step, dataSet), dataSet[0][0].map(_ => 0))

// linearRegression :: List Data -> List Number -> (Data -> Value)
function linearRegression(dataSet, step) {
  const theta = gradientDescent({ step, dataSet })
  return data => multiplyVectors(data, theta)
}



/* ----------------------------------------- *
        let's use it
* ----------------------------------------- */

const trainingSet = [
  {
    size: 2104,
    bedroomsCount: 3,
    floorsCount: 1,
    age: 8,
    price: 400
  },
  {
    size: 1416,
    bedroomsCount: 2,
    floorsCount: 1,
    age: 15,
    price: 232
  },
  {
    size: 1534,
    bedroomsCount: 3,
    floorsCount: 1,
    age: 2,
    price: 315
  },
  {
    size: 2500,
    bedroomsCount: 4,
    floorsCount: 2,
    age: 7,
    price: 460
  }
]

const formatData = ({ size, bedroomsCount, floorsCount, age }) =>
  [ 1, size, bedroomsCount, floorsCount, age ]

const formatTrainingSet = ({ price, ...rest }) => [
  formatData(rest),
  price
]

const h = linearRegression(trainingSet.map(formatTrainingSet), [.01, .0000001, .01, .01, .01])

console.log(`size: ${1503}, estimated price ${h(formatData({
  size: 1503,
  bedroomsCount: 1,
  floorsCount: 1,
  age: 4,
}))}`)

console.log(`size: ${2000}, estimated price ${h(formatData({
  size: 2000,
  bedroomsCount: 2,
  floorsCount: 2,
  age: 2,
}))}`)

console.log(`size: ${1503}, estimated price ${h(formatData({
  size: 1000,
  bedroomsCount: 3,
  floorsCount: 1,
  age: 18,
}))}`)