feature-scaling

This module allows you to feature scale an array of objects with numeric values.

You can install it with npm install feature-scaling or instead use the -g flag to install the global CLI.

For the objects it will treat the properties of the objects as stochastic variables. And project them with this formular:

x' = (x-min(X)) / (max(X) - min(X))

Values of the objects, that are not parseable as numbers will be ignored.

Example

var normalize = require('feature-scaling')

var data = [
  {a: 1, b:1},
  {a: "2", b:"1", note: "special"},
  {a: 3, b:10}
]

var normalized = normalize(data)

// normalized will be

// [ { a: 0,   b: 0 }, 
//   { a: 0.5, b: 0, note: 'special' }, 
//   { a: 1,   b: 1 } 
// ]

CLI usage

Just pipe in new line delimited JSON into feature-scaling command and it will return the data normalized. Note that this will load your entire input into memory. If this is an issue for you open an issue on this repo :).

Given data.ldjson:

  {"a": 1, "b":1}
  {"a": 2, "b":1}
  {"a": 3, "b":10}

This will output the following:

$ feature-scaling < data.ldjson
{"a":0,"b":0}
{"a":0.5,"b":0}
{"a":1,"b":1}