Skip to content

Sampler Sobol [JK03] [Owe95]

Description

The Sobol sequence. Samples are generated from a binary product between the binary representation of their index and a matrix, generated from primitive polynomials. By default, uses Juo & Kuo tables [JK03].

The sequence can be scrambled with Owen’s permutation [Owe95].

For a more precise description of this sampler and its performances in terms of aliasing and discrepancy, please refer to the following web bundle https://liris.cnrs.fr/ldbn/HTML_bundle/index.html.

Files

src/samplers/Sobol.cpp  
include/utk/samplers/SamplerSobol.hpp

Usage


Sobol sampler
Usage: ./Sobol [OPTIONS]

Options:
  -h,--help                   Print this help message and exit
  -n UINT REQUIRED            Number of points
  -d UINT REQUIRED            Dimensions
  -s,--seed UINT              Seed (unspecified means 'random')
  -m UINT [1]                 Number of pointsets
  -o,--out TEXT [out.dat]     Output file (format). {i} splits outputs in multiple files and token is replaced by index.
  --depth UINT [0]            Owen depth (0: no randomness, 32: recommended).
  --silent                    Silence UTK logs
  --table TEXT   Sobol init table file (Joe&Kuo format). If not specified  the [JK03] table is used.

#include <utk/utils/PointsetIO.hpp>
#include <utk/utils/Pointset.hpp>
#include <utk/samplers/SamplerStep.hpp>

int main()
{
    utk::Pointset<double> pts;

    // If needed, can add template to use 64bits integers
    utk::SamplerSobol sobol(2 /* dimension */, 32 /* depth */);
    sobol.setRandomSeed(args->seed);
    // Check for no errors
    if (sobol.generateSamples(pts, 1024 /* Number of points */))
    {
        write_text_pointset("sobol2d.dat", pts);
    }
}

import pyutk

# Only 32 bits integers supported
sobol = pyutk.Sobol(d=2, depth=32) 
samples = sobol.sample(1024) # This is a numpy array !

Results

License

See Licence.md file.