/WeightedRandomSelector

Very fast C# class for weighted random picking.

Primary LanguageC#MIT LicenseMIT

Weighted Random Selector

Description

This repository contains code for selecting items randomly based on weights. It is very commonly used thing in video game development and it is suprisingly tricky to get it right.

It was designed to be:

  • fast for smaller and bigger collections,
  • readable,
  • easy to use,
  • numerically precise
  • thread-safe selecting

Use cases

  • Item dropping from resources, enemies
  • Damage giving / taking
  • Selecting random events to happen
  • Procedural quest generation
  • Picking random sound effect to play on event
  • Casino games: Roulette, Slot machine (although I believe those casino games are weighted uniformly)
  • Random state machines / Markov chain

How to use

You have two options:

DynamicRandomSelector
  • Easier to use, since it contains both Construction and Random Picking parts
  • Can be modified on the fly (added or removed items), since internal collection is List
  • Around 2x slower, since internal collection is List
StaticRandomSelector
  • Harder to use
  • Separate component for Construction (StaticRandomSelectorBuilder) and RandomPicking (StaticRandomSelector)
  • Faster, since internal collection is Array
  • If you really need extra speed, remove all functions calls with hardcoding all functions

How it works?

Construction
  1. You input items, and every item has it's own weight (un-normalized)
  2. Array of weights is normalized, so that it becomes array of probabilities/masses (sum of 1)
  3. This array of probabilities get converted to (discrete) Cumulative Distribution Array
Random picking
  1. For each random pick, uniform random number is generated
  2. Depending on size of collection, linear or binary search is used to select correct index
  3. This index is used to return item from internal array/list.

Minimum of linear search and binary search

This picture demostrates what is happening under hood, to achieve good performance.

alt text

In short

Just use DynamicRandomSelector, and do not forget to call Build method after modifying (adding & removing) it.

Example code

var selector = new DynamicRandomSelector<float>();

// adding items
for (int i = 0; i < 32; i++) {

    var item = i;
    var unnormalizedWeight = Mathf.Sqrt(i + 1); //non zero weight

    selector.Add(item, unnormalizedWeight);
}

// building after modification is complete
selector.Build();

// picking random items
for (int i = 0; i < 10000; i++)
    var randomItem = selector.SelectRandomItem();