random-extra

random-extra

Seedable random number generator supporting many common distributions.

支援種子 (Seed) 的亂數產生器，提供多種常見分佈 (Distribution) 的抽樣函式。

this module fork from transitive-bullshit/random, with typescript support and some other change (include breaking change)

Welcome to the most random module on npm!

特色 (Highlights)

Wellcome send PR for more API support or performance up

• Simple API (make easy things easy and hard things possible)
• Seedable based on entropy or user input
• Plugin support for different pseudo random number generators (PRNGs)
• Sample from many common distributions
  • dfUniform, dfNormal, dfPoisson, dfBernoulli, etc, see the df prefix methods
• Validates all user input via chai
• Integrates with seedrandom
• Supports node.js >= 7 and browser (if here has no break)

breaking change: v2.x to v3.x

• for more easy know what api do by method name
• all (distribution function) method rename and add prefix df (ex: itemByWeight => dfItemByWeight) when run in loop, or wanna performance, pls use (distribution function) version
• remove shortid

安裝 (Installation)

npm install random-extra seedrandom

• benchmark 效能基準測試
• random.d.ts 型別宣告

使用方式 (Usage, 新版 API)

import random from 'random-extra';
import random = require('random-extra');

.use vs .newUse

• .use will change current random object
• .newUse will create new random object

設定 (preset)

seedrandom

use seedrandom for make seed-able

import seedrandom from '@lazy-random/preset-seedrandom';

注意 (Note)：preset-seedrandom 已拆分為獨立套件 @lazy-random/preset-seedrandom， 舊的 random-extra/preset/seedrandom 子路徑已不存在。 若不需要預設實例，也可改用下段「other way make seedrandom」的 random.newUse('seedrandom', ...) 寫法。

when use seedrandom, srand will able use

seedrandom.rand() // use current seed
seedrandom.srand() // every time call srand will make new seed
seedrandom.rand() // use new seed

other way make seedrandom

import random from 'random-extra';
const seedrandom = random.newUse('seedrandom')

import _seedrandom = require('seedrandom')

random.newUse(_seedrandom('hello.', { entropy: true }))
random.newUse(_seedrandom('hello.', { entropy: false }))

random.newUse(_seedrandom('hello.'))

使用方式 (Usage, 主要 API)

const random = require('random-extra')

// quick uniform shortcuts
random.float(min = 0, max = 1) // uniform float in [ min, max )
random.int(min = 0, max = 1) // uniform integer in [ min, max ]
random.boolean() // true or false

// uniform
random.dfUniform(min = 0, max = 1) // () => [ min, max )
random.dfUniformInt(min = 0, max = 1) // () => [ min, max ]
random.dfUniformBoolean() // () => [ false, true ]

// normal
random.dfNormal(mu = 0, sigma = 1)
random.dfLogNormal(mu = 0, sigma = 1)

// bernoulli
random.dfBernoulli(p = 0.5)
random.dfBinomial(n = 1, p = 0.5)
random.dfGeometric(p = 0.5)

// poisson
random.dfPoisson(lambda = 1)
random.dfExponential(lambda = 1)

// misc
random.dfIrwinHall(n)
random.dfBates(n)
random.dfPareto(alpha)

For convenience, several common dfUniform samplers are exposed directly:

random.float()     // 0.2149383367670885
random.int(0, 100) // 72
random.boolean()   // true

All distribution methods return a thunk (function with no params), which will return a series of independent, identically distributed random variables from the specified distribution.

// create a normal distribution with default params (mu=1 and sigma=0)
const normal = random.dfNormal()
normal() // 0.4855465422678824
normal() // -0.06696771815439678
normal() // 0.7350852689834705

// create a poisson distribution with default params (lambda=1)
const poisson = random.dfPoisson()
poisson() // 0
poisson() // 4
poisson() // 1

Note that returning a thunk here is more efficient when generating multiple samples from the same distribution.

You can change the underlying PRNG or its seed as follows:

const seedrandom = require('seedrandom')

// change the underlying pseudo random number generator
// by default, Math.random is used as the underlying PRNG
random.use(seedrandom('foobar'))

// create a new independent random number generator
const rng = random.clone('my-new-seed')

// create a second independent random number generator and use a seeded PRNG
const rng2 = random.clone(seedrandom('kittyfoo'))

API

Table of Contents

• Random

Random

Seedable random number generator supporting many common distributions.

Defaults to Math.random as its underlying pseudorandom number generator.

Type: function (rng)

• rng (Rng | function) Underlying pseudorandom number generator. (optional, default Math.random)

---

Random 實例方法 (Instance Methods)

• clone(seed?, opts?, ...args?) — 建立一個新的 Random 實例，可指定新種子 (Seed)。
• use(rng) — 替換目前實例的底層亂數產生器 (PRNG)，並回傳自身，可串接使用。
• newUse(rng, ...args) — 以指定的亂數產生器另外建立一個新的 Random 實例。
• cloneUse(rng, ...args) — 先 clone 出新實例，再對新實例執行 use。

// use：就地替換目前的 PRNG
random.use('xor128', 'foobar')

// newUse：保留目前實例，回傳新的實例
const r2 = random.newUse('seedrandom', 'hello.', null)

// cloneUse：複製後再換 PRNG
const r3 = random.cloneUse('seedrandom', 'kittyfoo')

分佈函式 (Distribution Functions)

所有 df 前綴的分佈函式皆回傳一個 thunk（無參數函式），重複呼叫可高效取得同一分佈的連續亂數樣本：

分佈 (Distribution)	方法
均勻分佈 (Uniform)	dfUniform、dfUniformInt、dfUniformBoolean
常態分佈 (Normal)	dfNormal、dfLogNormal
伯努利分佈 (Bernoulli)	dfBernoulli、dfBinomial、dfGeometric
泊松分佈 (Poisson)	dfPoisson、dfExponential
其他 (Misc)	dfIrwinHall、dfBates、dfPareto
陣列／權重 (Array / Weighted)	dfItemByWeight、dfItemByWeightUnique、dfArrayShuffle、dfArrayUnique、dfSumInt、dfSumFloat 等

完整清單請見 src/random.d.ts 與 RandomCore 的型別宣告。

Todo

• Distributions

  • dfUniform
  • dfUniformInt
  • dfUniformBoolean
  • dfNormal
  • dfLogNormal
  • chiSquared
  • cauchy
  • fischerF
  • studentT
  • dfBernoulli
  • dfBinomial
  • negativeBinomial
  • dfGeometric
  • dfPoisson
  • dfExponential
  • gamma
  • hyperExponential
  • weibull
  • beta
  • laplace
  • dfIrwinHall
  • dfBates
  • dfPareto

• Generators

  • pluggable prng
  • port more prng from boost
  • custom entropy

• Misc

  • browser support via rollup
  • basic docs
  • basic tests
  • full test suite
  • initial release!

開發 (Development)

在 monorepo 根目錄或本套件目錄下執行：

pnpm run test
pnpm run test:jest
pnpm run build:tsc
pnpm run benchmark

變更日誌 (Changelog)

請見 CHANGELOG.md。

相關資源 (Related)

• d3-random - D3's excellent random number generation library.
• seedrandom - Seedable pseudo random number generator.
• random-int - For the common use case of generating dfUniform random ints.
• random-float - For the common use case of generating dfUniform random floats.
• randombytes - Random crypto bytes for Node.js and the browser.

Credit

Huge shoutout to Roger Combs for donating the random npm package for this project!

Lots of inspiration from d3-random (@mbostock and @svanschooten).

Some distributions and PRNGs are ported from C++ boost::random.

License

MIT Travis Fischer