Random Numbers & Distributions in D — mir-random Guide

How to generate random numbers and sample distributions with mir-random. For the full API, browse mir-random.libmir.org.

All examples compile against current mir-random. mir-random separates engines (the bit source) from variables (distributions that consume an engine) — pick an engine, then draw from any distribution with it. Note: mir-random's PRNGs are statistically sound but not cryptographically secure.

Reproducible engine + uniform draws

import mir.random;
import mir.random.variable;
import mir.random.engine.xorshift;

auto gen = Xorshift(1234);                 // explicit seed → reproducible
auto u = uniformVar!double(0, 1)(gen);     // uniform in [0, 1)
auto i = uniformVar!int(0, 6)(gen);        // integer in [0, 6]

Seed an engine explicitly when you need reproducible streams (tests, simulations). uniformVar!T(lo, hi) builds a distribution object; calling it with an engine draws a value.

Choosing an engine

import mir.random.variable;
import mir.random.engine.xorshift;
import mir.random.engine.mersenne_twister;

auto fast = Xorshift(1234);        // small, fast, good general-purpose
auto mt   = Mt19937_64(42);        // Mersenne Twister, long period
auto v    = uniformVar!double(-5, 5)(mt);

Xorshift is a compact, fast default; Mt19937_64 gives the familiar Mersenne Twister with a very long period. mir-random also ships PCG and xoshiro engines — see mir.random.engine.

Sampling distributions

import mir.random.variable;
import mir.random.engine.xorshift;

auto gen = Xorshift(1234);
auto n = normalVar!double(0, 1)(gen);      // standard normal (mean 0, sd 1)
auto e = exponentialVar!double(1.5)(gen);  // exponential
auto g = gammaVar!double(2.0)(gen);        // gamma, shape = 2

Each *Var constructor returns a reusable distribution object — build it once, draw many times by calling it with an engine. The full catalogue (Bernoulli, Binomial, Beta, Cauchy, Poisson, …) is in mir.random.variable.

Where to go next

• mir.random.variable — every distribution
• mir.random.engine — every PRNG engine
• mir.random.ndvariable — multivariate sampling
• ndslice Cookbook — fill arrays and process the results