libRmath.js

Javascript Pure Implementation of Statistical R "core" numerical libRmath.so

R-js

470

TypeScript

libRmath.js

This R statistical nmath re-created in typescript/javascript.

This is the documentation of 2.0.0, for the lts branch of version 1.x see here.

If you were not using a previous version to 2.0.0, you can skip breaking changes and go to:

Installation and usage
Table of contents

BREAKING CHANGES For version 2.0

Removed

RNG (normal and uniform) are only selectable via `RNGkind` function.

The normal and uniform implementation of the various RNG’s are not exported publicly anymore
Select normal and uniform RNG’s via the function RNGkind.

// this is NOT possible anymore
import { AhrensDieter } from 'lib-r-math.js';
const ad = new AhrensDieter();
ad.random();

// NEW way of doing things
import { RNGkind, rnorm } from 'lib-r-math.js';
RNGkind({ normal: 'AHRENS_DIETER' }); // R analog to "RNGkind"
rnorm(8); // get 8 samples, if you only want one sample consider rnormOne()

helper functions for data mangling

Functions removed from 2.0.0 onwards: any, arrayrify, multiplex, each, flatten, c, map, selector, seq, summary.

It is recommended you either use well established js tools like Rxjs or Ramdajs to mangle arrays and data.

Removed helper functions for limiting numeric precision

Functions removed from 2.0.0 onwards: numberPrecision

This function mimicked the R’s options(digits=N).

Changed

helper functions

Functions changed from 2.0.0 onwards: timeseed.

timeseed is now replaced by a cryptographic safe seed seed.

Sample distributions return a result of type `Float64Array`.

Functions changed from 2.0.0 onwards:

All these functions will return type of Float64Array:
rbeta, rbinom, rcauchy, rchisq, rexp, rf, rgamma, rgeom, rhyper, rlogis, rlnorm, rmultinom, rnorm, rpois, rsignrank, rt,runif, rweibull, rwilcox.

For single scalar (number) return values, use the analogs:
rbetaOne, rbinomOne, rcauchyOne, rchisqOne, rexpOne, rfOne, rgammaOne, rgeomOne, rhyperOne, rlogisOne, rlnormOne, rnormOne, rpoisOne, rsignrankOne, rtOne,runifOne, rweibullOne, rwilcoxOne.

Example:

import { rbinom, rbinomOne, setSeed } from 'lib-r-math.js';

rbinom(0); //
// -> FloatArray(0)

setSeed(123); // set.seed(123) in R
rbinom(2, 8, 0.5);
// -> Float64Array(2) [ 3, 5 ]  //same result as in R

setSeed(456); // set.seed(456) in R
rbinomOne(350, 0.5);
// -> 174  ( a single scalar )

UMD module removed

There is no UMD module from 2.0.0. These are the module types for node and browser

node: esm (multiple files), commonjs
web: esm (single bundle) and iife (“immediately-invoked Function Expression”)

Installation and usage

Minimal version of node required is 16.18.1.

npm i lib-r-math.js

lib-r-math.js supports the following module types:

ESM for use in observablehq

library = import('https://cdn.skypack.dev/lib-r-math.js@latest/dist/web.esm.mjs');
library.BesselJ(3, 0.4);
//-> -0.30192051329163955

ESM for use as Browser client

<script type="module">
    import { BesselJ } from 'https://unpkg.dev/lib-r-math.js@latest/dist/web.esm.mjs';

    console.log(BesselJ(3, 0.4));
    //-> -0.30192051329163955
</script>

IIFE (immediately-invoked Function Expression) for use in Browser client

<script src="https://unpkg.dev/lib-r-math.js@latest/dist/web.iife.js"></script>
<script>
    const answ = window.R.BesselJ(3, 0.4);
    console.log(answ);
    //-> -0.30192051329163955
</script>

ESM for Node

import { BesselJ } from 'lib-r-math.js';

const answ = BesselJ(3, 0.4);
//-> -0.30192051329163955

COMMONJS for node

const { BesselJ } = require('lib-r-math.js');

const answ = BesselJ(3, 0.4);
//-> -0.30192051329163955

libRmath.js

Auxiliary functions

`RNGkind`

RNGkind is the analog to R’s “RNGkind”. This is how you select what RNG (normal and uniform) you use and the samplingKind

Follows closely the R implementation here

R console:

> RNGkind()
[1] "Mersenne-Twister" "Ahrens-Dieter"
[3] "Rejection"

Just like in R, calling RNGkind with no argument returns the currently active RNG’s (uniform and normal) and sample kind (Rounding or Rejection)

Like in R, RNGkind optionally takes an argument of type RandomGenSet, after processing it will return the (adjusted) RandomGenSet indicating what RNG’s and “kind of sampling” is being used.

Rjs typescript decl:

function RNGkind(options?: RandomGenSet): RandomGenSet;

Arguments:

options: an object of type RandomGenSet
- options.uniform: string, specify name of uniform RNG to use.
- options.normal: string, specify nam of normal RNG (shaper) to use
- options.sampleKind: string, specify sample strategy to use

Typescript definition:

type RandomGenSet = {
    uniform?:
        | 'KNUTH_TAOCP'
        | 'KNUTH_TAOCP2002'
        | 'LECUYER_CMRG'
        | 'MARSAGLIA_MULTICARRY'
        | 'MERSENNE_TWISTER'
        | 'SUPER_DUPER'
        | 'WICHMANN_HILL';
    normal?: 'AHRENS_DIETER' | 'BOX_MULLER' | 'BUGGY_KINDERMAN_RAMAGE' | 'KINDERMAN_RAMAGE' | 'INVERSION';
    sampleKind?: 'ROUNDING' | 'REJECTION';
};

The RNGkind function is decorated with the following extra properties:

property	description	example
`RNGkind.uniform`	list of constants of uniform RNG’s	`RNGkind.uniform.MARSAGLIA_MULTICARRY` is equal to the string `"MARSAGLIA_MULTICARRY"`
`RNGkind.normal`	list of constants of normal RNG’s	`RNGkind.normal.KINDERMAN_RAMAGE` is equal to the string `"KINDERMAN_RAMAGE"`
`RNGkind.sampleKind`	list of sampling strategies	`RNGkind.sampleKind.ROUNDING` is equal to the string `"ROUNDING"`

Example: set uniform RNG to SUPER_DUPER and normal RNG to BOX_MULLER

import { RNGkind } from 'lib-r-math.js';

const uniform = RNGkind.uniform.SUPER_DUPER;
const normal = RNGkind.normal.BOX_MULLER;

RNGkind({ uniform, normal }); //-> "sampleKind" not specified so this will not be changed

RNGkind(); // no arguments, will return the current used RNG's and "sampleKind"
// returns
//  {
//    uniform: 'SUPER_DUPER',
//    normal: 'BOX_MULLER',
//    sampleKind: 'ROUNDING'  // was not changed from default setting
//  }

`setSeed`

Uses a single value to initialize the internal state of the currently selected uniform RNG.

R console analog: set.seed

Rjs typescript decl

function setSeed(s: number): void;

Arguments:

s is coerced to an unsigned 32 bit integer

`randomSeed`

R console analog: .Random.seed

Rjs typescript decl

function randomSeed(internalState?: Uint32Array | Int32Array): Uint32Array | Int32Array | never;

Arguments:

(optional) internalState: the value of a previously saved RNG state, the current RNG state will be set to this.
return state of the current selected RNG

Exceptions:

If the internalState value is not correct for the RNG selected an Error will be thrown.

Distributions

All distribution functions follow a prefix pattern:

d (like dbeta, dgamma) are density functions
p (like pbeta, pgamma) are (cumulative) distribution function
q (like qbeta, qgamma) are quantile functions
r (like rbeta/rbetaOne, rgamma/rgammaOne) generates random deviates

The Beta distribution

type	function spec
density function	`function dbeta(x: number, shape1: number, shape2: number, ncp?: number, log = false): number`
distribution function	`function pbeta(q: number, shape1: number, shape2: number, ncp?: number, lowerTail = true, logP = false): number`
quantile function	`function qbeta(p: number, shape1: number, shape2: number, ncp?: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rbeta(n: number, shape1: number, shape2: number, ncp?: number): Float32Array`
random generation	`function rbetaOne(shape1: number, shape2: number): number`

Arguments:
- x, q: quantile value
- p: probability
- n: number of observations
- shape1, shape2: Shape parameters of the Beta distribution
- log, logP: if true, probabilities are given as log(p).
- lowerTail: if true, probabilities are P[X ≤ x], otherwise, P[X > x].

Example:

import { dbeta } from 'lib-r-math.js';

dbeta(0.5, 2, 2);
// -> 1.5

The Binomial distribution

type	function spec
density function	`function dbinom(x: number, n: number, prob: number, log = false): number`
distribution function	`function pbinom(q: number, n: number, prob: number, lowerTail = true, logP = false): number`
quantile function	`function qbinom(p: number, size: number, prob: number, lower_tail = true, logP = false): number`
random generation (bulk)	`function rbinom(n: number, size: number, prob: number): Float64Array`
random generation	`function rbinomOne(size: number, prob: number): number`

Arguments:
- x, q: quantile value
- p: probability
- n: number of observations.
- size: number of trials (zero or more).
- prob: probability of success on each trial.
- log, logP: if true, probabilities are given as log(p).
- lowerTail: if true, probabilities are P[X ≤ x], otherwise, P[X > x].

Example:

import { dbinom } from 'lib-r-math.js';

dbinom(50, 100, 0.5);
// -> 0.07958924

The Negative Binomial Distribution

type	function spec
density function	`function dnbinom(x: number, size: number, prob?: number, mu?: number, log = false): number`
distribution function	`function pnbinom(q: number, size: number, prob?: number, mu?: number, lowerTail = true, logP = false): number`
quantile function	`function qnbinom(p: number, size: number, prob?: number, mu?: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rnbinom(n: number, size: number, prob?: number, mu?: number): Float64Array`
random generation	`function rnbinom(size: number, prob?: number, mu?: number): number`

Arguments:

x, q: quantile value.
p: probability.
n: number of observations.
size: target for number of successful trials, (need not be integer) or dispersion parameter (the shape parameter of the gamma mixing distribution). Must be strictly positive.
prob: probability of success in each trial. 0 < prob <= 1.
mu: alternative parametrization via mean: see ‘Details’.
log, logP: if true, probabilities are given as log(p).
lowerTail: if true, probabilities are P[X ≤ x], otherwise, P[X > x].

Details: R doc

A negative binomial distribution can also arise as a mixture of Poisson distributions with mean distributed as a gamma distribution (see pgamma) with scale parameter (1 - prob)/prob and shape parameter size. (This definition allows non-integer values of size.)

An alternative parametrization (often used in ecology) is by the mean mu (see above), and size, the dispersion parameter, where prob = size/(size+mu). The variance is mu + mu^2/size in this parametrization.

Example:

R console:

> options(digits=22)
> 126 /  dnbinom(0:8, size  = 2, prob  = 1/2)
[1]   504.0000000000000000000   503.9999999999998863132   672.0000000000000000000  1008.0000000000001136868
[5]  1612.7999999999994997779  2688.0000000000013642421  4607.9999999999972715159  8064.0000000000000000000
[9] 14336.0000000000145519152

Equivalence in js (fidelity):

import { dnbinom } from 'lib-r-math.js';

console.log([0, 1, 2, 3, 4, 5, 6, 7, 8].map((x) => 126 / dnbinom(x, 2, 0.5)));
// ->
[
    504, 503.9999999999999, 672, 1008.0000000000001, 1612.7999999999988, 2688.0000000000014, 4607.999999999997, 8064,
    14336.000000000015
];

The Cauchy Distribution

type	function spec
density function	`function dcauchy(x: number, location = 0, scale = 1, log = false): number`
distribution function	`function pcauchy(x: number, location = 0, scale = 1, lowerTail = true, logP = false): number`
quantile function	`function qcauchy(p: number, location = 0, scale = 1, lowerTail = true, logP = false): number`
random generation (bulk)	`function rcauchy(n: number, location = 0, scale = 1): Float32Array`
random generation	`function rcauchyOne(location = 0, scale = 1): number`

Arguments:

x, q: quantile value.
p: probability.
n: number of observations.
location, scale: location and scale parameters.
log, logP: if true, probabilities are given as log(p).
lowerTail: if true, probabilities are P[X ≤ x], otherwise, P[X > x].

Examples

R console:

dcauchy(-1:4)
[1] 0.15915494309189534560822 0.31830988618379069121644 0.15915494309189534560822 0.06366197723675813546773
[5] 0.03183098861837906773387 0.01872411095198768526959

Equivalence in js (fidelity):

import { dcauchy } from 'lib-r-math.js';

console.log([-1, 0, 1, 2, 3, 4].map((x) => dcauchy(x)));
// -> [  0.15915494309189535, 0.3183098861837907, 0.15915494309189535, 0.06366197723675814, 0.03183098861837907, 0.018724110951987685 ]

The Chi-Squared (non-central) Distribution

type	function spec
density function	`function dchisq(x: number, df: number, ncp?: number, log = false ): number`
distribution function	`function pchisq(p: number, df: number, ncp?: number, lowerTail = true, logP = false ): number`
quantile function	`function qchisq(p: number, df: number, ncp?: number, lowerTail = true, logP = false ): number`
random generation (bulk)	`function rchisq(n: number, df: number, ncp?: number): Float64Array`
random generation	`function rchisqOne(df: number, ncp?: number): number`

Arguments:

x, q: quantile.
p: probability.
n: number of observations.
df: degrees of freedom (non-negative, but can be non-integer).
ncp: non-centrality parameter (non-negative).
log, logP: if true, probabily p are given as log§.
lowerTail: if true`TRUE (default), probabilities are P[X \le x]P[X≤x], otherwise, P[X > x]P[X>x].

Examples

R console:

dchisq(1, df = 1:3)
[1] 0.2419707 0.3032653 0.2419707

Equivalence in js (fidelity):

import { dchisq } from 'lib-r-math.js';

console.log([1, 2, 3].map((df) => dchisq(1, df)));
// -> [ 0.24197072451914337, 0.3032653298563167, 0.24197072451914337 ]

The Exponential Distribution

type	function spec
density function	`function dexp(x: number, rate = 1, log = false): number`
distribution function	`function pexp(q: number, rate = 1, lowerTail = true, logP = false): number`
quantile function	`function qexp(p: number, rate = 1, lowerTail = true, logP = false): number`
random generation (bulk)	`function rexp(n: number, rate = 1):Float64Array`
random generation	`function rexpOne(rate = 1): number`

Arguments:

x, q: quantile.
p: probabily.
n: number of observations.
rate: the exp rate parameter
log, logP: if true, probabilities p are given as log(p).
lower.tail: if true (default), probabilities are P[ X ≤ x ], otherwise, P[X > x].

Examples

R console:

dexp(1) - exp(-1)
[1] 0

Equivalence in js (fidelity):

import { dexp } from 'lib-r-math.js';

console.log(dexp(1) - Math.exp(-1));
// -> 0

The F Distribution

type	function spec
density function	`function df(x: number, df1: number, df2: number, ncp?: number, log = false): number`
distribution function	`function pf(q: number, df1: number, df2: number, ncp?: number, lowerTail = true, logP = false): number`
quantile function	`function qf(p: number, df1: number, df2: number, ncp?: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rf(n: number, df1: number, df2: number, ncp?: number): Float64Array`
random generation	`function rfOne(df1: number, df2: number, ncp?: number): number`

Arguments:

x, q: quantile.
p: probabily.
n: number of observations.
df1, df1: degrees of freedom. Infinity is allowed.
ncp: non-centrality parameter. If omitted the central F is assumed.
log, logP: if true, probabilities p are given as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x ], otherwise, P[X > x]S.

**NOTE: JS has no named arguments for functions, so specify ncp = undefined, if you want to change the log, logP, lowerTail away from their defaults

Examples

R console:

## Identity (F <-> Beta <-> incompl.beta):
n1 <- 7 ; n2 <- 12; qF <- c((0:4)/4, 1.5, 2:16)
x <- n2/(n2 + n1*qF)
stopifnot(all.equal(pf(qF, n1, n2, lower.tail=FALSE),
                    pbeta(x, n2/2, n1/2)))

Equivalence in js (fidelity):

import { pf, pbeta } from 'lib-r-math.js';

var qF = [
    0.0, 0.25, 0.5, 0.75, 1.0, 1.5, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0
];

var n1 = 7;
var n2 = 12;
var xs = qF.map((qf) => n2 / (n2 + n1 * qf));

var betas = xs.map((x) => pbeta(x, n2 / 2, n1 / 2));

var fisher = qF.map((qf) => pf(qf, n1, n2, undefined /*no ncp*/, false));

// array "betas" and "fisher" should be equal

console.log(fisher.map((f, i) => f - betas[i]));
//-> [ 0, 0, 0, 0, 0, ...., 0]

The Gamma Distribution

type	function spec
density function	`function dgamma(x: number, shape: number, rate?: number, scale?: number, log = false): number`
distribution function	`function pgamma(q: number, shape: number, rate?: number, scale?: number, lowerTail = true, logP = false): number`
quantile function	`function qgamma(p: number, shape: number, rate?: number, scale?: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rgamma(n: number, shape: number, rate?: number, scale?: number): Float64Array`
random generation	`function rgammaOne(shape: number, rate?: number, scale?: number): number`

Arguments:

x, q: quantile
p: probability
n: number of observations.
rate: an alternative way to specify the scale.
shape, scale: shape and scale parameters. Must be positive, scale strictly.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

-log(dgamma(1:4, shape = 1))
[1] 1 2 3 4

Equivalence in js (fidelity):

import { dgamma } from 'lib-r-math.js';

let dg = [1, 2, 3, 4].map((x) => Math.log(dgamma(x, 1)));
// -> [ -1, -2, -3, -4 ]
//
// this is Equivalence to to
// [1,2,3,4].map (x => dgamma(x, 1, undefined, undefined, true) );

The Geometric Distribution

type	function spec
density function	`function dgeom(x: number, p: number, log = false): number`
distribution function	`function qgeom(p: number, prob: number, lowerTail = true, logP = false): number`
quantile function	`function qgeom(p: number, prob: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rgeom(n: number, prob: number): Float64Array`
random generation	`function rgeomOne(p: number): number`

Arguments:

x, q: quantile
p: probability
n: number of observations.
prob: probability of success in each trial. 0 < prob <= 1.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

qgeom((1:9)/10, prob = .2)
[1]  0  0  1  2  3  4  5  7 10

Equivalence in js (fidelity):

import { qgeom } from 'lib-r-math.js';

let dg = [1, 2, 3, 4, 5, 6, 7, 8, 9].map((p) => p / 10).map((p) => qgeom(p, 0.2));

console.log(dg);
// -> [ 0, 0, 1,  2, 3, 4, 5, 7, 10 ]

The Hypergeometric Distribution (Web Assembly accalerated)

type	function spec
density function	`function dhyper(x: number, m: number, n: number, k: number, log = false): number`
distribution function	`function phyper(q: number, m: number, n: number, k: number, lowerTail = true, logP = false): number`
quantile function	`function qhyper(p: number, m: number, n: number, k: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rhyper(nn: number, m: number, n: number, k: number): Float64Array`
random generation	`function rhyperOne(m: number, n: number, k: number): number`

Arguments:

x, q: quantile
p: probability
m: the number of white balls in the urn.
n: the number of black balls in the urn.
k: the number of balls drawn from the urn, hence must be in 0,1,…,m+n.
p: probability, it must be between 0 and 1.
nn: number of observations.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

m <- 10; n <- 7; k <- 8
x <- 0:(k+1)
rbind(phyper(x, m, n, k), dhyper(x, m, n, k))
     [,1]         [,2]       [,3]     [,4]      [,5]      [,6]      [,7]       [,8]       [,9] [,10]
[1,]    0 0.0004113534 0.01336898 0.117030 0.4193747 0.7821884 0.9635952 0.99814891 1.00000000     1
[2,]    0 0.0004113534 0.01295763 0.103661 0.3023447 0.3628137 0.1814068 0.03455368 0.00185109     0

Equivalence in js (fidelity):

import { phyper, dhyper } from "lib-r-math.js";
var m = 10;
var n = 7;
var k = 8;
var xs = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9];

console.log( ...xs.map( x => phyper(x, m, n, k)));
0 0.000411353352529823 0.01336898395721926 0.11703002879473474 0.4193747429041546 0.7821883998354585 0.9635952283011107 0.9981489099136158 1 1

console.log( ...xs.map( x => dhyper(x, m, n, k)));
0 0.000411353352529823 0.012957630604689437 0.10366104483751548 0.30234471410941993 0.3628136569313041 0.18140682846565206 0.03455368161250514 0.001851090086384205 0

Web Assembly backend

Use useWasmBackendHyperGeom and clearBackendHyperGeom to enable/disable wasm backend.

import {
    useWasmBackendHyperGeom,
    clearBackendHyperGeom,
    //
    qhyper
} from 'lib-r-math.js';

// the functions "qhyper" will be accelerated (on part with native C for node >=16)
useWasmBackendHyperGeom();

qhyper(0.5, 2 ** 31 - 1, 2 ** 31 - 1, 2 ** 31 - 1); // 28 sec (4.3 Ghz Pentium) wasm big numbers to make it do some work
// -> 1073741806

clearBackendHyperGeom(); // revert to js backend

qhyper(0.5, 2 ** 31 - 1, 2 ** 31 - 1, 2 ** 31 - 1); // this will take 428 sec (4.3 Ghz Pentium)
// -> 1073741806

The Logistic Distribution

type	function spec
density function	`function dlogis(x: number, location = 0, scale = 1, log = false): number`
distribution function	`function plogis(x: number, location = 0, scale = 1, lowerTail = true, logP = false): number`
quantile function	`function qlogis(p: number, location = 0, scale = 1, lowerTail = true, logP = false): number`
random generation (bulk)	`function rlogis(n: number, location = 0, scale = 1): Float64Array`
random generation	`function rlogisOne(location = 0, scale = 1): number`

Arguments:

x, q: quantile
p: probability
location, scale: location and scale parameters.
n: number of observations.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

> RNGkind()
[1] "Mersenne-Twister" "Inversion"        "Rejection"
> set.seed(12345)
> var(rlogis(4000, 0, scale = 5))
[1] 80.83207

Equivalence in js (fidelity):

import { setSeed, RNGkind, rlogis } from 'lib-r-math.js';

const uniform = RNGkind.uniform.MERSENNE_TWISTER;
const normal = RNGkind.normal.INVERSION;

RNGkind({ uniform, normal });
setSeed(12345);

let samples = rlogis(4000, 0, 5); // get 4000 samples

// calculate sample variance
const N = samples.length;
const µ = samples.reduce((sum, x) => sum + x, 0) / N;
const S = (1 / (N - 1)) * samples.reduce((sum, x) => sum + (x - µ) ** 2, 0); // sample variance

console.log(S);
// -> 80.83207248898108 (fidelity proven)

The Log Normal Distribution

type	function spec
density function	`function dlnorm(x: number, meanlog = 0, sdlog = 1, log = false): number`
distribution function	`function plnorm(q: number, meanlog = 0, sdlog = 1, lowerTail = true, logP = false): number`
quantile function	`function qlnorm(p: number, meanlog = 0, sdlog = 1, lowerTail = true, logP = false): number`
random generation (bulk)	`function rlnorm(n: number, meanlog = 0, sdlog = 1): Float32Array`
random generation	`function rlnormOne(meanlog = 0, sdlog = 1): number`

Arguments:

x, q: quantile
p: probability
meanlog, sdlog: mean and standard deviation of the distribution on the log scale with default values of 0 and 1 respectively.
n: number of observations.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Examples:

R console:

dlnorm(1) == dnorm(0)
[1] TRUE

Equivalence in js (fidelity):

import { dlnorm, dnorm } from 'lib-r-math.js';
console.log(dlnorm(1) === dnorm(0));
// -> true

The Multinomial Distribution

type	function spec
density function	`function dmultinom(x: Float32Array, prob: Float32Array, log = false): number`
density function (R like)	`function dmultinomLikeR(x: Float32Array, prob: Float32Array, log = false): number`
random generation (bulk)	`function rmultinom(n: number, size: number, prob: Float64Array): Float64Array`

Arguments:

x: quantile
n: number of random vectors to draw.
size:
- integer, say N, specifying the total number of objects that are put into K boxes in the typical multinomial experiment.
- dmultinom omit’s the size parameter (used in R version), see “Details” below for motivation.
prob: numeric non-negative array of length K, specifying the probability for the K classes; is internally normalized to sum 1. Infinite and missing values are not allowed.
log: if true, log probabilities are computed.

Motivation for removing size argument from dmultinom:

The code snippet shows clarification

N <- sum(x)
if (is.null(size)) # if size is the default (null) then assign it the value N (number of )
  size <- N
else if (size != N) # if manually set AND not equal to sum(x) throw Error,
  stop("size != sum(x), i.e. one is wrong")

Because of the above R code allowing manual setting of size in dmultinom is omitted

Example:

R console:

> RNGkind()
[1] "Mersenne-Twister" "Inversion"        "Rejection"
> set.seed(1234)
> rmultinom(10, size = 12, prob = c(0.1,0.2,0.8))
     [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
[1,]    0    1    2    0    1    1    0    0    0     0
[2,]    3    3    2    1    2    2    4    4    1     1
[3,]    9    8    8   11    9    9    8    8   11    11
>

Equivalence in js (fidelity):

import { RNGkind, setSeed, rmultinom } from 'lib-r-math.js';

RNGkind({
    uniform: RNGkind.uniform.MERSENNE_TWISTER,
    normal: RNGkind.normal.INVERSION
});

setSeed(1234); // use same seed as in R example

const answer = rmultinom(10, 12, new Float64Array([0.1, 0.2, 0.8]));
// returns a (row-first) matrix as a single Float64Array with size (prob.length x size)

console.log(...answer);
// -> 0 3 9 1 3 8 2 2 8 0 1 11 1 2 9 1 2 9 0 4 8 0 4 8 0 1 11 0 1 11
// first column 0 3 9
// second column 1 3 8  etc etc

The Normal Distribution

type	function spec
density function	`function dnorm(x: number, mean = 0, sd = 1, log = false): number`
distribution function	`function pnorm(q: number, mean = 0, sd = 1, lowerTail = true, logP = false): number`
quantile function	`function qnorm(p: number, mean = 0, sd = 1, lowerTail = true, logP = false): number`
random generation (bulk)	`function rnorm(n: number, mean = 0, sd = 1): Float64Array`
random generation	`function rnormOne(mean = 0, sd = 1): number`

Arguments:

x, q: quantile
p: probability
mean, sd: mean and standard deviation.
n: number of observations.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

dnorm(0) == 1/sqrt(2*pi)
[1] TRUE
dnorm(1) == exp(-1/2)/sqrt(2*pi)
[1] TRUE
dnorm(1) == 1/sqrt(2*pi*exp(1))
[1] TRUE

Equivalence in js:

import { dnorm } from 'lib-r-math.js';

const { sqrt, exp, PI: pi } = Math;

console.log(dnorm(1) === exp(-1 / 2) / sqrt(2 * pi));
// -> true
console.log(dnorm(1) === exp(-1 / 2) / sqrt(2 * pi));
// -> true
console.log(dnorm(1) === 1 / sqrt(2 * pi * exp(1)));
// -> true

The Poisson distribution

type	function spec
density function	`function dpois(x: number, lambda: number, log = false): number`
distribution function	`function ppois(q: number,lambda: number, lowerTail = true, logP = false): number`
quantile function	`function qpois(p: number, lambda: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rpois(n: number, lamda: number): Float32Array`
random generation	`function rpoisOne(lambda: number): number`

Arguments:

x, q: quantile.
p: probability.
lambda: non-negative mean.
n: number of observations.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

> options(digits=20)
> -log(dpois(0:7, lambda = 1) * gamma(1+ 0:7)) # == 1
[1] 1.00000000000000000000 1.00000000000000000000 1.00000000000000000000 1.00000000000000000000
[5] 0.99999999999999977796 1.00000000000000022204 1.00000000000000022204 1.00000000000000000000

Equivalence in js:

import { dpois, gamma } from 'lib-r-math.js';

const { log } = Math;
let arr = [0, 1, 2, 3, 4, 5, 6, 7];
let result = arr.map((x) => -log(dpois(x, 1) * gamma(x + 1)));
console.log(...result);
//-> 1 1 1 1 0.9999999999999996 1 1.0000000000000009 0.9999999999999989

Distribution of the Wilcoxon Signed Rank Statistic

type	function spec
density function	`function dsignrank(x: number, n: number, log = false): number`
distribution function	`function psignrank(q: number, n: number, lowerTail = true, logP = false): number`
quantile function	`function qsignrank(p: number, n: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rsignrank(nn: number, n: number): Float64Array`
random generation	`function rsignrank(nn, n): number`

Arguments:

x, q: quantile.
p: probability.
nn: number of observations.
n: number of observations in the sample. A positive integer.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Examples:

R console:

> options(digits=20)
> x=seq(0,5*6/2)
> y=dsignrank(x, 5)
> data.frame(x,y)
    x                       y
    0 0.031250000000000000000
    1 0.031250000000000000000
    2 0.031250000000000000000
    3 0.062500000000000000000
    4 0.062500000000000000000
    5 0.093749999999999986122
    6 0.093749999999999986122
    7 0.093749999999999986122
    8 0.093749999999999986122
    9 0.093749999999999986122
   10 0.093749999999999986122
   11 0.062500000000000000000
   12 0.062500000000000000000
   13 0.031250000000000000000
   14 0.031250000000000000000
   15 0.031250000000000000000

Equivalence in js:

import { dsignrank } from 'lib-r-math.js';
const N = 5;
for (let x = 0; x <= (N * (N + 1)) / 2; x++) {
    console.log(x, dsignrank(x, N));
}
/*
0 0.03125
1 0.03125
2 0.03125
3 0.0625
4 0.0625
5 0.09374999999999999
6 0.09374999999999999
7 0.09374999999999999
8 0.09374999999999999
9 0.09374999999999999
10 0.09374999999999999
11 0.0625
12 0.0625
13 0.03125
14 0.03125
15 0.03125
*/

Output As Graphic:

Web Assembly backend

dsignrank, psignrank and qsignrank have an optional Web Assembly backend, turn this backend on/off with useWasmBackendSignRank and clearBackendSignRank respectivily.

Example

import { useWasmBackendSignRank, clearBackendSignRank, psignrank } from 'lib-r-math.js';

useWasmBackendSignRank();
// all sign rank functions accelerated

const p = psignrank(...); // so something usefull

clearBackendSignRank();
// use javascript backend for signrank distribution

The Student t Distribution

type	function spec
density function	`function dt(x: number, df: number, ncp = 0, log = false): number`
distribution function	`function pt(q: number, df: number, ncp = 0, lowerTail = true, logP = false): number`
quantile function	`function qt(p: number, df: number, ncp?: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rt(n: number, df: number, ncp?: number): Float64Array`
random generation	`function rtOne(df: number): number`

Arguments:

x, q: quantile.
p: probability.
n: number of observations.
df: degrees of freedom (>0, maybe non-integer). df = Inf is allowed.
ncp: non-centrality parameter $\delta$; currently except for rt(), only for abs(ncp) <= 37.62. If omitted, use the central t distribution.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

1 - pt(1:5, df = 1)
[1] 0.2499999999999998
[2] 0.1475836176504333
[3] 0.1024163823495667
[4] 0.0779791303773694
[5] 0.0628329581890011

Equivalence in js:

import { pt } from 'lib-r-math.js';

for (let q = 1; q <= 5; q++) {
    console.log(1 - pt(q, 1));
}
// 0.24999998762491238
// 0.14758361679076415
// 0.10241638219629579
// 0.07797913031910642
// 0.06283295806783729

The Studentized Range Distribution

type	function spec
distribution function	`function ptukey(q: number, nmeans: number, df: number, nrnages = 1, lowerTail = true, logP = false): number`
quantile function	`function qt(p: number, df: number, ncp?: number, lowerTail = true, logP = false): number`

Arguments:

q: quantile.
p: probability.
nmeans: sample size for range (same for each group).
df: degrees of freedom for ss (see below).
nranges: number of groups whose maximum range is considered.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

 ptukey(-1:8, nm = 6, df = 5)
 [1] 0.0000000000000000 0.0000000000000000 0.0272115020859732
 [4] 0.2779845061609432 0.6007971569446733 0.8017143642776676
 [7] 0.9014257065957741 0.9489495069295981 0.9721701726664311
[10] 0.9840420193770625

Equivalence in js:

import { ptukey } from 'lib-r-math.js';

function* generatePTukeyData() {
    for (let q = -1; q <= 8; q++) {
        yield ptukey(q, 6, 5);
    }
}

console.log(...generatePTukeyData());
// 0                  0                  0.02721150208597321
// 0.2779845061609432 0.6007971569446733 0.8017143642776676
// 0.9014257065957741 0.9489495069295981 0.9721701726664311
// 0.9840420193770627

Output As Graphic:

The Uniform Distribution

type	function spec
density function	`function dunif(x: number, min = 0, max = 1, log = false): number`
distribution function	`function punif(q: number, min = 0, max = 1, lowerTail = true, logP = false): number`
quantile function	`function qunif(p: number, min = 0, max = 1, lowerTail = true, logP = false): number`
random generation (bulk)	`function runif(n: number, min = 0, max = 1): Float64Array`
random generation	`function runifOne(min: number, max: number): number`

Arguments:

x,q: quantile.
p: probability.
min, max: lower and upper limits of the distribution. Must be finite.
n: number of observations.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

> RNGkind()
[1] "Mersenne-Twister" "Inversion"        "Rejection"
> set.seed(12345)
> runif(5)
[1] 0.720903896261007 0.875773193081841 0.760982328327373 0.886124566197395
[5] 0.456480960128829

Equivalence in js:

import { RNGkind, setSeed, runif } from 'lib-r-math.js';

// these are defaults so skip setting via "RNGkind" this set if not changed

const uniform = RNGkind.uniform.MERSENNE_TWISTER;
const normal = RNGkind.normal.INVERSION;
RNGkind({ uniform, normal });

// set seed
setSeed(12345);
console.log(...runif(5));
// -> 0.7209038962610066 0.8757731930818409 0.7609823283273727 0.8861245661973953 0.4564809601288289

The Weibull Distribution

type	function spec
density function	`function dweibull(x: number, shape: number, scale = 1, log = false): number`
distribution function	`function pweibull(q: number, shape: number, scale = 1, lowerTail = true, logP = false): number`
quantile function	`function qweibull(p: number, shape: number, scale = 1, lowerTail = true, logP = false): number`
random generation (bulk)	`function rweibull(n: number, shape: number, scale = 1): Float64Array`
random generation	`function rweibullOne(shape: number, scale = 1): number`

Arguments:

x,q: quantile.
p: probability.
n: number of observations.
shape, scale: shape and scale parameters, the latter defaulting to 1.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

x <- c(0, rlnorm(50))
all.equal(dweibull(x, shape = 1), dexp(x))

Equivalence in js:

import { rlnorm, dweibull, dexp } from 'lib-r-math.js';

const samples = rlnorm(50);

const violation = samples.find((x) => dweibull(x, 1) !== dexp(x));

console.log(violation);
// -> undefined, no violation!

Distribution of the Wilcoxon Rank Sum Statistic

type	function spec
density function	`function dwilcox(x: number, m: number, n: number, log = false): number`
distribution function	`function pwilcox(q: number, m: number, n: number, lowerTail = true, logP = false): number`
quantile function	`function qwilcox(x: number, m: number, n: number, lowerTail = true, logP = false): number`
random generation (bulk)	`function rwilcox(nn: number, m: number, n: number): Float32Array`
random generation	`function rwilcoxOne(m: number, n: number): number`

Arguments:

x, q: quantile.
p: probability.
nn: number of observations.
m, n: numbers of observations in the first and second sample, respectively.
log, logP: if true, probabilities/densities p are returned as log(p).
lowerTail: if true (default), probabilities are P[ X ≤ x], otherwise, P[X > x].

Example:

R console:

> x <- seq(-1, (4*6 + 1), 4);
> fx = dwilcox(x,4,6)
> fx
[1] 0.00000000000000000 0.01428571428571429 0.04761904761904762 0.07619047619047620 0.06666666666666667
[6] 0.02857142857142857 0.00476190476190476

Equivalence in js:

import { dwilcox } from 'lib-r-math.js';

for (let x = -1; x <= 4 * 6 + 1; x += 4) {
    console.log(dwilcox(x, 4, 6));
}
// ->
// 0
// 0.014285714285714285
// 0.047619047619047616
// 0.0761904761904762
// 0.06666666666666667
// 0.02857142857142857
// 0.004761904761904762

Special Functions of Mathematics

Special functions are particular mathematical functions which have more or less established names and notations due to their importance in mathematical analysis, functional analysis, physics, or other applications.

There is no general formal definition, but the list of mathematical functions contains functions which are commonly accepted as special.

Bessel functions

Bessel Functions of integer and fractional order, of first and second kind, $J_{\nu}$ and $Y_{\nu}$, and Modified Bessel functions (of first and third kind), $I_{\nu}$ and $K_{\nu}$.

type	function spec
Modified Bessel function of the first kind $I_{\nu}$	`function BesselI(x: number, nu: number, exponScaled = false): number`
Modified Bessel function of the third kind $K_{\nu}$	`function BesselK(x: number, nu: number, exponScaled = false): number`
Bessel function of the first kind $J_{\nu}$	`function BesselJ(x: number, nu: number): number`
Bessel function of the second kind $Y_{\nu}$	`function BesselY(x: number, nu: number): number`

Arguments:

x: must be ≥ 0.
nu: The order (maybe fractional and negative) of the corresponding Bessel function.
exponScaled: if true, the results are exponentially scaled in order to avoid overflow ( $I_{\nu}$ ) or underflow ( $K_{\nu}$ ), respectively.

Details:
If exponScaled = true, $e^{-x} \cdot I_{\nu}(x)$ or $e^{x} \cdot K_{\nu}(x)$ are returned.

For $\nu < 0$, formulae 9.1.2 and 9.6.2 from Abramowitz & Stegun are applied (which is probably suboptimal), except for besselK which is symmetric in nu.

The current algorithms will give warnings about accuracy loss for large arguments. In some cases, these warnings are exaggerated, and the precision is perfect. For large nu, say in the order of millions, the current algorithms are rarely useful.

Example:

R console:

> data.frame(besselI(0, 0:4), besselI(1, 0:4), besselI(2, 0:4), besselI(3, 0:4), besselI(4, 0:4))
  besselI.0..0.4.     besselI.1..0.4.    besselI.2..0.4.   besselI.3..0.4.   besselI.4..0.4.
1               1 1.26606587775200818 2.2795853023360673 4.880792585865024 11.30192195213633
2               0 0.56515910399248503 1.5906368546373291 3.953370217402609  9.75946515370445
3               0 0.13574766976703828 0.6889484476987382 2.245212440929951  6.42218937528411
4               0 0.02216842492433190 0.2127399592398526 0.959753629496008  3.33727577842035
5               0 0.00273712022104687 0.0507285699791802 0.325705181937935  1.41627570765359

Equivalence in js:

import { BesselI } from 'lib-r-math.js';

for (let nu = 0; nu <= 4; nu++) {
    const row = [0, 1, 2, 3, 4].map((x) => BesselI(x, nu) + '\t');
    console.log(...row);
}
/*
1        1.2660658777520082      2.2795853023360673      4.880792585865024       11.301921952136333
0        0.565159103992485       1.590636854637329       3.9533702174026093      9.75946515370445
0        0.13574766976703828     0.6889484476987382      2.245212440929951       6.422189375284106
0        0.022168424924331902    0.21273995923985264     0.959753629496008       3.337275778420345
0        0.002737120221046866    0.05072856997918024     0.32570518193793546     1.4162757076535895
*/

Beta functions

The functions beta and lbeta return the beta function and the natural logarithm of the beta function,

type	function spec
beta function	`function beta(a: number, b: number): number`
logarithem of the beta function	`function lbeta(a: number, b: number): number`

Arguments:

a, b: non negative values quantile.

The formal definition is

$$ B(a, b) = \int_0^1 t^{a-1} (1-t)^{b-1} dt $$

(Abramowitz and Stegun section 6.2.1, page 258). Note that it is only defined in R for non-negative a and b, and is infinite if either is zero.

No examples provided, usage is straightforward

Gamma functions

type	function spec
gamma function $\Gamma(x) = \int_{0}^{\infty}t{x-1}e^{-t}dt$	`function gamma(x: number): number`
natural logarithm of the $\Gamma(x)$ function	`function lgamma(x: number, sgn?: Int32Array)`

Arguments:

x: number, can be negative.
sgn: (optional) an array, if provided, will have its first element set to the sign of x

Polygamma functions

type	function spec
first derivative of the logarithm of the gamma function $\Psi_{0}(x) = \frac{\Gamma^{\prime}(x)}{\Gamma(x)}$	`function digamma(x: number): number`
second derivative of the logarithm of the gamma function $\Psi_{1}(x) = \frac{d^2}{dx2}ln\Gamma(x)$	`function trigamma(x: number): number`
third derivative of the logarithm of the gamma function $\Psi_{2}(x) = \frac{d^3}{dx3}ln\Gamma(x)$	`function tetragamma(x: number): number`
forth derivative of the logarithm of the gamma function $\Psi_{3}(x) = \frac{d^4}{dx4}ln\Gamma(x)$	`function pentagamma(x: number): number`
Nth derivative of the logarithm of the gamma function $\Psi_{n-1}(x) = \frac{d^n}{dxn}ln\Gamma(x)$	`function psigamma(x: number, deriv: number): number`

Arguments:

x: number, can be negative.
deriv: >= 0,
- deriv = 0 computes the first derivative $\Psi_{0}(x) = \frac{\Gamma^{\prime}(x)}{\Gamma(x)}$
- deriv = 1 computes the second derivative of the logarithm of the gamma function $\Psi_{1}(x) = \frac{d^2}{dx2}ln\Gamma(x)$
- deriv = N computes the (N+1)th derivative of the logarithm of the gamma function $\Psi_{n}(x) = \frac{d^n}{dxn}ln\Gamma(x)$

Binomial coefficient functions

type	function spec
binomial coefficient ${n}\choose{k}$	`function choose(n: number, k: number): number`
natural log of $\left\|{n}\choose{k}\right\|$	`function lchoose(n: number, k: number): number`

Arguments:

n,k: “n over k”.
For $ k \ge 1 $ it is defined as $ \frac{n(n-1)\cdots(n-k+1)}{k!} $.
For $k=0$ it is defined as 1.
For $k \lt 0$, it is defined as 0.

libRmath.js

libRmath.js

BREAKING CHANGES For version 2.0

Removed

RNG (normal and uniform) are only selectable via RNGkind function.

helper functions for data mangling

Removed helper functions for limiting numeric precision

Changed

helper functions

Sample distributions return a result of type Float64Array.

UMD module removed

Installation and usage

ESM for use in observablehq

ESM for use as Browser client

IIFE (immediately-invoked Function Expression) for use in Browser client

ESM for Node

COMMONJS for node

Table of Contents

Auxiliary functions

RNGkind

setSeed

randomSeed

Distributions

The Beta distribution

The Binomial distribution

The Negative Binomial Distribution

The Cauchy Distribution

The Chi-Squared (non-central) Distribution

The Exponential Distribution

The F Distribution

The Gamma Distribution

The Geometric Distribution

The Hypergeometric Distribution (Web Assembly accalerated)

Web Assembly backend

The Logistic Distribution

The Log Normal Distribution

The Multinomial Distribution

The Normal Distribution

The Poisson distribution

Distribution of the Wilcoxon Signed Rank Statistic

Web Assembly backend

The Student t Distribution

The Studentized Range Distribution

The Uniform Distribution

The Weibull Distribution

Distribution of the Wilcoxon Rank Sum Statistic

Special Functions of Mathematics

Bessel functions

Beta functions

Gamma functions

Polygamma functions

Binomial coefficient functions

RNG (normal and uniform) are only selectable via `RNGkind` function.

Sample distributions return a result of type `Float64Array`.

`RNGkind`

`setSeed`

`randomSeed`