Package ovh.paulem.arcana.math

Class FastRandom

java.lang.Object

ovh.paulem.arcana.math.FastRandom

All Implemented Interfaces:

Serializable

public class FastRandom
extends Object
implements Serializable

An RNG class that's faster and hopefully better than java.util.Random.

(1) Is NOT synchronized! Thus NOT thread-safe. Therefore faster. In most applications I can think of, every thread should get its own RNG. Threadsafe can be nice for convenient utility functions; use java.util.Random for that. (2) Uses the XORShift RNG algorithm, which is faster and more accurate (more random) than the algorithm in java.util.Random. (3) Folds in the high-level methods from Mallet's cc.mallet.util.Randoms (4) Leaves out some of the convenience methods that appear in java.util.Random but not Mallet.

XORShift implementation from ... Mallet methods from ... ... i have no idea what license status that leaves this file

See discussions e.g. ... ...

Speed tests, compared to java.util.Random - see main(): - 10.7 times faster for nextUniform() - 1.8 times faster for nextInt(1000) - I swapped this into an LDA collapsed Gibbs sampler, and got a 5% speedup of the entire program

Looking for feedback / bug reports.

See Also:

• Serialized Form

Field Summary

Fields

Modifier and Type	Field	Description
static final FastRandom	INSTANCE

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	FastRandom()	Creates a new RNG and seeds it using the default seeding strategy.
	FastRandom(byte[] seed)	Creates an RNG and seeds it with the specified seed data.
protected	FastRandom(long seed)

Method Summary

Modifier and Type	Method	Description
static int	convertBytesToInt(byte[] bytes, int offset)	Take four bytes from the specified position in the specified block and convert them into a 32-bit int, using the big-endian convention.
static int[]	convertBytesToInts(byte[] bytes)	Convert an array of bytes into an array of ints. 4 bytes from the input data map to a single int in the output data.
static FastRandom	getJ17FastRandom()
byte[]	getSeed()
static void	main(String[] args)
protected int	next(int bits)
double	nextBeta(double alpha, double beta)	Return a random double drawn from a Beta distribution with mean a/(a+b) and variance ab/((a+b+1)(a+b)^2).
BitSet	nextBitSet(int size, double p)	Return a random BitSet with "size" bits, each having probability p of being true.
boolean	nextBoolean()	Return a random boolean, equally likely to be true or false.
boolean	nextBoolean(double p)	Return a random boolean, with probability p of being true.
void	nextBytes(byte[] bytes)
double	nextChiSq()	Return a random double drawn from an Chi-squarted distribution with mean 1 and variance 2.
double	nextChiSq(int df)	Return a random double drawn from an Chi-squared distribution with mean df and variance 2*df.
double	nextChiSq(int df, double lambda)	Return a random double drawn from an Chi-squared distribution with mean df+lambda and variance 2*df.
int	nextDiscrete(double[] a)	Draw a single sample from multinomial "a".
int	nextDiscrete(double[] a, double sum)	draw a single sample from (unnormalized) multinomial "a", with normalizing factor "sum".
double	nextDouble()
double	nextExp()	Return a random double drawn from an Exponential distribution with mean 1 and variance 1.
double	nextExp(double beta)	Return a random double drawn from an Exponential distribution with mean beta and variance beta^2.
double	nextExp(double beta, double lambda)	Return a random double drawn from an Exponential distribution with mean beta+lambda and variance beta^2.
float	nextFloat()
double	nextGamma()	Return a random double drawn from a Gamma distribution with mean 1.0 and variance 1.0.
double	nextGamma(double alpha)	Return a random double drawn from a Gamma distribution with mean alpha and variance 1.0.
double	nextGamma(double alpha, double beta)	Return a random double drawn from a Gamma distribution with mean alpha*beta and variance alpha*beta^2.
double	nextGamma(double alpha, double beta, double lambda)	Return a random double drawn from a Gamma distribution with mean alpha*beta+lamba and variance alpha*beta^2.
double	nextGaussian()	Return a random double drawn from a Gaussian distribution with mean 0 and variance 1.
double	nextGaussian(double m, double s2)	Return a random double drawn from a Gaussian distribution with mean m and variance s2.
int	nextInt()
int	nextInt(int n)
long	nextLong()
int	nextPoisson()	Return nextPoisson(1).
int	nextPoisson(double lambda)	Return random integer from Poission with parameter lambda.
double	nextUniform()	Return a random double in the range 0 to 1, inclusive, uniformly sampled from that range.
double	nextUniform(double a, double b)	Return a random double in the range a to b, inclusive, uniformly sampled from that range.
static FastRandom	of()
double	oldNextGamma(int ia)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

INSTANCE

public static final FastRandom INSTANCE

Constructor Details

FastRandom

protected FastRandom()

Creates a new RNG and seeds it using the default seeding strategy.

FastRandom

protected FastRandom(long seed)

FastRandom

public FastRandom(byte[] seed)

Creates an RNG and seeds it with the specified seed data.

Parameters:

seed - The seed data used to initialise the RNG.

Method Details

of

public static FastRandom of()

getJ17FastRandom

public static FastRandom getJ17FastRandom()

convertBytesToInt

public static int convertBytesToInt(byte[] bytes,
 int offset)

Take four bytes from the specified position in the specified block and convert them into a 32-bit int, using the big-endian convention.

Parameters:

bytes - The data to read from.

offset - The position to start reading the 4-byte int from.

Returns:

The 32-bit integer represented by the four bytes.

convertBytesToInts

public static int[] convertBytesToInts(byte[] bytes)

Convert an array of bytes into an array of ints. 4 bytes from the input data map to a single int in the output data.

Parameters:

bytes - The data to read from.

Returns:

An array of 32-bit integers constructed from the data.

Since:

1.1

main

public static void main(String[] args)

getSeed

public byte[] getSeed()

next

protected int next(int bits)

nextInt

public int nextInt(int n)

nextPoisson

public int nextPoisson(double lambda)

Return random integer from Poission with parameter lambda. The mean of this distribution is lambda. The variance is lambda.

nextPoisson

public int nextPoisson()

Return nextPoisson(1).

nextBoolean

public boolean nextBoolean()

Return a random boolean, equally likely to be true or false.

nextBoolean

public boolean nextBoolean(double p)

Return a random boolean, with probability p of being true.

nextBitSet

public BitSet nextBitSet(int size,
 double p)

Return a random BitSet with "size" bits, each having probability p of being true.

nextUniform

public double nextUniform()

Return a random double in the range 0 to 1, inclusive, uniformly sampled from that range. The mean of this distribution is 0.5. The variance is 1/12.

nextUniform

public double nextUniform(double a,
 double b)

Return a random double in the range a to b, inclusive, uniformly sampled from that range. The mean of this distribution is (b-a)/2. The variance is (b-a)^2/12

nextDiscrete

public int nextDiscrete(double[] a)

Draw a single sample from multinomial "a".

nextDiscrete

public int nextDiscrete(double[] a,
 double sum)

draw a single sample from (unnormalized) multinomial "a", with normalizing factor "sum".

nextGaussian

public double nextGaussian()

Return a random double drawn from a Gaussian distribution with mean 0 and variance 1.

nextGaussian

public double nextGaussian(double m,
 double s2)

Return a random double drawn from a Gaussian distribution with mean m and variance s2.

nextGamma

public double nextGamma()

Return a random double drawn from a Gamma distribution with mean 1.0 and variance 1.0.

nextGamma

public double nextGamma(double alpha)

Return a random double drawn from a Gamma distribution with mean alpha and variance 1.0.

oldNextGamma

public double oldNextGamma(int ia)

nextGamma

public double nextGamma(double alpha,
 double beta)

Return a random double drawn from a Gamma distribution with mean alpha*beta and variance alpha*beta^2.

nextGamma

public double nextGamma(double alpha,
 double beta,
 double lambda)

Return a random double drawn from a Gamma distribution with mean alpha*beta+lamba and variance alpha*beta^2. Note that this means the pdf is: frac{ x^{alpha-1} exp(-x/beta) }{ beta^alpha Gamma(alpha) } in other words, beta is a "scale" parameter. An alternative parameterization would use 1/beta, the "rate" parameter.

nextExp

public double nextExp()

Return a random double drawn from an Exponential distribution with mean 1 and variance 1.

nextExp

public double nextExp(double beta)

Return a random double drawn from an Exponential distribution with mean beta and variance beta^2.

nextExp

public double nextExp(double beta,
 double lambda)

Return a random double drawn from an Exponential distribution with mean beta+lambda and variance beta^2.

nextChiSq

public double nextChiSq()

Return a random double drawn from an Chi-squarted distribution with mean 1 and variance 2. Equivalent to nextChiSq(1)

nextChiSq

public double nextChiSq(int df)

Return a random double drawn from an Chi-squared distribution with mean df and variance 2*df.

nextChiSq

public double nextChiSq(int df,
 double lambda)

Return a random double drawn from an Chi-squared distribution with mean df+lambda and variance 2*df.

nextBeta

public double nextBeta(double alpha,
 double beta)

Return a random double drawn from a Beta distribution with mean a/(a+b) and variance ab/((a+b+1)(a+b)^2).

nextInt

public int nextInt()

nextLong

public long nextLong()

nextBytes

public void nextBytes(byte[] bytes)

nextDouble

public double nextDouble()

nextFloat

public float nextFloat()