weighted

A dead-simple module for picking an item from a set of items while picking some more frequently than others. Each item is given a numerical "weight": each item's likelihood to be selected is directly proportional to its share of the total weight.

For example, if I have two items, "apple" and "orange", with "apple" having a weight of 1 and "orange" having a weight of 2, then "orange" is twice as likely to be chosen as "apple" (2/3 vs 1/3). If we add a third option, "banana", with a weight of 2, then "banana" and "orange" are each as likely to be picked as the other, and twice as likely to be picked as "apple" (2/5 vs 1/5).

Installation

npm install weighted

Usage

Weighted returns a Function additionally available as weighted.select:

weighted(set, weights, [options])
weighted(obj, [options])
weighted.select(set, weights, [options])
weighted.select(obj, [options])

Selections can be made based on either a pair of Arrays (one with the items, and one with their weights) or a single Object (with the items as keys and their weights as values). In each of the following examples, 'Wake Up' has a 25% chance of being selected, while 'Snooze Alarm' has a 75% chance. If we added more items, the chances would change accordingly as discussed above.

Array version

var weighted = require('weighted')

var items = ['Wake Up', 'Snooze Alarm']
  , weights = [0.25, 0.75]

console.log('Decision:', weighted.select(items, weights))

Object version

var weighted = require('weighted')

var items = {
      'Wake Up': 0.25,
      'Snooze Alarm': 0.75
    }

console.log('Decision:', weighted.select(items))

Available options

The third argument, options, shapes the way Weighted performs the selection:

Name	Description
rand	A function returning a number in [0, ) to use for selection. Assumed to be (though by no means required) uniformly distributed. See "Overriding rand", below.
normal	If truthy all weights are assumed to add up to 1. See "Pre-normalizing for performance", below.

Overriding rand

By default Weighted uses Math.random() for selection. This can be overridden by providing a Function as the rand option. Like Math.random(), it should require no arguments and return a number between 0 and 1. For example, if you wanted to override rand with the mersenne library:

var weighted = require('weighted')
  , mersenne = require('mersenne')

var items = ['Wake Up', 'Snooze Alarm']
  , weights = [0.25, 0.75]

function rand() {
  return mersenne.rand() / 32768
}

console.log('Decision:', weighted.select(items, weights, rand))

Pre-normalizing

This module is optimized for multiple selections from the same set of items, and will try to front-load the costs of normalizing any weights you assign them. but if the set of items or their weights need to change constantly, you can help Weighted out by normalizing or tallying those weights yourself.

If all the weights add up to 1 (as they have in our examples), you can pass a truthy value to the normal option, and Weighted will assume as much. If all the weights add up to any other value, you can provide a number as the total option, and Weighted will use that total instead.

From a human perspective, it can be convenient to normalize your weights, as any weight multiplied by 100 will give you the exact percentage likelihood that item will be selected. On the other hand, if there is some "voting" mechanism influencing the weights, for instance, simply using the number of votes as the weight for each item will work well enough.

Microbenchmarks are available via npm run bench if you want examples of the different patterns and their performance implications.

Selecting multiple items simultaneously

For now, selecting multiple items in the same call is unsupported. If you want multiple selections (i.e. duplicates are okay, you just want multiple), then calling weighted.select multiple times is sufficient. If you're hoping to get multiple unique selections in the same call, you're best off writing a wrapper to suit your specific use case. Issue #2 has one approach.

Thanks

This module was originally built for bot-factory, and I'd be remiss if I didn't thank Red Robot Labs for sponsoring that work, and the AI team at 38 Studios for teaching me just how "intelligent" random activity can seem.