public class HaltonSequenceGenerator extends Object implements RandomVectorGenerator
A Halton sequence is a low-discrepancy sequence generating points in the interval [0, 1] according to
H(n) = d_0 / b + d_1 / b^2 .... d_j / b^j+1 with n = d_j * b^j-1 + ... d_1 * b + d_0 * b^0For higher dimensions, subsequent prime numbers are used as base, e.g. { 2, 3, 5 } for a Halton sequence in R^3.
Halton sequences are known to suffer from linear correlation for larger prime numbers, thus the individual digits are usually scrambled. This implementation already comes with support for up to 40 dimensions with optimal weight numbers from H. Chi: Scrambled quasirandom sequences and their applications.
The generator supports two modes:
nextVector()
skipTo(int)
Constructor and Description |
---|
HaltonSequenceGenerator(int dimension)
Construct a new Halton sequence generator for the given space dimension.
|
HaltonSequenceGenerator(int dimension,
int[] bases,
int[] weights)
Construct a new Halton sequence generator with the given base numbers and weights for each dimension.
|
Modifier and Type | Method and Description |
---|---|
int |
getNextIndex()
Returns the index i of the next point in the Halton sequence that will be returned
by calling
nextVector() . |
double[] |
nextVector()
Generate a random vector.
|
protected int |
scramble(int i,
int j,
int b,
int digit)
Performs scrambling of digit
d_j according to the formula: |
double[] |
skipTo(int index)
Skip to the i-th point in the Halton sequence.
|
public HaltonSequenceGenerator(int dimension) throws OutOfRangeException
dimension
- the space dimensionOutOfRangeException
- if the space dimension is outside the allowed range of [1, 40]public HaltonSequenceGenerator(int dimension, int[] bases, int[] weights) throws NullArgumentException, OutOfRangeException, DimensionMismatchException
dimension
- the space dimensionbases
- the base number for each dimension, entries should be (pairwise) prime, may not be nullweights
- the weights used during scrambling, may be null in which case no scrambling will be performedNullArgumentException
- if base is nullOutOfRangeException
- if the space dimension is outside the range [1, len], where
len refers to the length of the bases arrayDimensionMismatchException
- if weights is non-null and the length of the input arrays differpublic double[] nextVector()
nextVector
in interface RandomVectorGenerator
protected int scramble(int i, int j, int b, int digit)
d_j
according to the formula:
( weight_i * d_j ) mod baseImplementations can override this method to do a different scrambling.
i
- the dimension indexj
- the digit indexb
- the base for this dimensiondigit
- the j-th digitpublic double[] skipTo(int index) throws NotPositiveException
This operation can be performed in O(1).
index
- the index in the sequence to skip toNotPositiveException
- if index < 0public int getNextIndex()
nextVector()
.Copyright © 2003–2016 The Apache Software Foundation. All rights reserved.