Sankhya

Sankhya, a JavaScript utility-library to declaratively transform an Object into another using pure transformation functions.

This is inspired from Plumbing Clojure library

Installation

npm i sankhya-graph-js
or
yarn add sankhya-graph-js

Example

Traditional way of defining Transformation:

function stats(input) {
  const {values} = input

  if (!values) throw new Error('No property "values"!')

  const count = values.length
  const mean = values.reduce(sum) / count
  const meanSquare = values.map(square).reduce(sum) / count
  const variance = meanSquare - square(mean)

  const output = {
    count,
    mean,
    meanSquare,
    variance,
  }
  return output
}

function sum(a, b) {
  return a + b
}
function square(a) {
  return a * a
}

const data = {values: [1, 2, 3, 4, 5, 6, 7]}
const transformedObj = stats(data)

console.log(transformedObj)
// -> Object {count: 7, mean: 4, meanSquare: 20, variance: 4}

Problems with the above approach:

Here the drawback of this approach is that each of these transformation function are dependent on the ones above it. Hence we have to make sure that proper dependency order is maintained.

So if you try to exchange the order of the definition of these functions it will blow up. If you exchange meanSquare and variance order, it won't work.

Here we can see that these functions are not independent which becomes a liability

Hidden Graph which is maintained by this stats

Using sankhya, we can refactor stats to this very declarative, robust and structured form, using a map from keywords to keyword functions:

const sankhya = require('sankhya-graph-js')

const square = x => x ** 2
const sum = (x, y) => x + y

const stats = sankhya({
  meanSquare: (i, o) => i.values.map(square).reduce(sum) / o.count,
  count: (i, o) => i.values.length,
  variance: (i, o) => o.meanSquare - square(o.mean),
  mean: (i, o) => i.values.reduce(sum) / o.count,
})

const transformedObj = stats({values: [1, 2, 3, 4, 5, 6, 7]})
console.log(transformedObj)
// -> Object {count: 7, meanSquare: 20, mean: 4, variance: 4}

Explanation:

Every output transformation is pure and expressed with an arrow function, where the i and o parameters represent input (data) and output (the future returned object) respectively.

We can express output values in function of other output values, sankhya will understand the underlying graph and execute the statements in the right order! Ordering of the definition of these functions can be arbitrary

Moreover, each micro-function in the object values is executed exacly once, and the result cached for next calls. (You can verify this by logging from inside the arrow functions)

This is why the given functions need to be pure.

Validation

Before performing any transformation sankhya validates input data, and throws if a property on input is non-existant.

function dataAttributeProxy(data) {
  return new Proxy(data, {
    get: (t, p, r) => {
      if (p in t) {
        return t[p]
      }
      throw new Error(`Data object is missing key ${p}`)
    },
  })
}

Laziness

Sometimes we define a heavy computation function, rich of all the data we might need, but somewhere we need only a subset of these.

But in the traditional way we can't achieve this without actually triggering all of the computations

For these cases we have .lazy.

For example, suppose we only need the mean of our data:

const data = {values: [1, 2, 3, 4, 5, 6, 7]}
const transformed = stats.lazy(data)

// The object is initially empty:
console.log(transformed)
// -> Object {}

// We materialize and objectify the computations as they are needed:
console.log(transformed.mean)
// -> 4

// The results are then attached to the object,
// but anything we don't need is not computed:

console.log(transformed)
// -> Object {count: 7, mean: 4}

This is made possible by using prototype getters.

This also guarantees to execute the minimum of necessary steps to get the result.