[Python] Problems with repeated draws of Sobol sequence in scipy

How do I get unique (uncorrelated) Sobol sequences by making repeated calls to random or random_base2?

I'm trying to repeatedly generate 'random' Sobol sequences, but I'm getting the same sequence each time. My reading of the documentation suggested that repeated calls would give me a different sequence ("To continue an existing design, extra points can be obtained by calling again random_base2.").

Here's my code:

import scipy
engine2 = scipy.stats.qmc.Sobol(d=1, scramble=True)
sample_qmc1 = engine2.random_base2(m=14)
print(sample_qmc1)
sample_qmc2 = engine2.random_base2(m=14)
print(sample_qmc2)


Here's example output:

[[0.2534843 ]
[0.884218 ]
[0.69113704]
...
[0.69106254]
[0.88420477]
[0.25347106]]
[[0.25342032]
[0.88416122]
[0.69107275]
...
[0.69111674]
[0.88426601]
[0.2535251 ]]


As you can see, the values are almost the same.

reset() doesn't help, I get the same sequence. Changing to random() doesn't help either. Using smaller sample sizes does seem to work.

If I add this line:

sample_qmc3 = engine2.random_base2(m=14)


I get this error:

ValueError: The balance properties of Sobol' points require n to be a power of 2. 24576 points have been previously generated, then: n=24576+2**14=40960. If you still want to do this, the function 'Sobol.random()' can be used.


I tried LatinHypercube and it doesn't have this problem.

I'm using scipy 1.14.1

Anyone go any ideas how I can generate unique Sobol sequences?

Continue reading...