System.Random.Dice

This module implements an n-sided dice and provides sampling from a given integer range. The algorithm uses rejection sampling and attempts to keep the total number of used random bits as close as possible to the information theoretic lower bound of ln(n) / ln(2) (for a range of size n).

The implementation exposes streams of random values as conduits, see diceRolls and randomRs. We also provide IO wrappers around these functions, see getDiceRolls and getRandomRs. The conduit interface allows us to use a specific entropy source, which has type Producer IO Word.

Usage:

If we wanted to use the system-specific entropy source (systemEntropy) to produce 10 dice rolls of a 6-sided dice (i.e. range [0,5]), we could write:

> systemEntropy $$ diceRolls 6 =$= CL.take 10
[5,1,3,3,0,5,3,2,2,1]

The function testPerformance yields the actual number of consumed random bits:

> testPerformance 12 10000
Generated 10000 random samples in range [0,11]
Average number of bits used: 3.5904
Entropy lower bound on the number of required bits: 3.5849625007211565
Performance ratio: 1.0015167520658164

Documentation

diceRolls :: Int -> Conduit Word8 IO Int

Produces a stream of random integer values in the range [0,n-1], for a given n <= 2^55. This conduit needs to be attached to an entropy source such as systemEntropy.

randomRs

Arguments

:: (Int, Int)

range (inclusive)

-> Conduit Word8 IO Int

Produces a stream of random integer values within a range. This conduit needs to be attached to an entropy source such as systemEntropy.

getDiceRolls

Arguments

:: Int

n: number of sides

-> Int

k: number of rolls

-> IO [Int]

Generates k rolls of an n sided dice.

getRandomRs

Arguments

:: (Int, Int)

(inclusive) range

-> Int

number of samples

-> IO [Int]

Generates a list of random integer values in the specified range.

systemEntropy :: Producer IO Word8

A source of entropy. By default, we use the getEntropy function from the entropy package, see systemEntropy.

Warning: When combining a source of entropy with other conduits, it is important that there is no "backflow" due to leftover values that are being returned to the source from the conduit. This can be done by fusing the conduit with the identity map, e.g: myEntropySrc $$ Data.Conduit.List.map id =$= myConduit