Python Importing Fashion Affects the Sampler Output's Shape

[ehsansaleh]

Describe the bug
There are two ways to call the generate_samples function:

• Calling generate_samples with a from chaospy.distributions.sampler.generator import generate_samples import.
• Calling sampler.generator.generate_samples with a from chaospy.distributions import sampler import.

As you can see below, these two methods yield different output shapes on the same set of input arguments.

To Reproduce

>>> import chaospy
>>> from chaospy.distributions.sampler.generator import generate_samples
>>> from chaospy.distributions import sampler

>>> cpy_dist = chaospy.Iid(chaospy.Normal(0, 1), 1)
>>> cpy_jac = chaospy.J(cpy_dist)
>>> rule = "additive_recursion"

>>> a = generate_samples(order=128, domain=cpy_jac, rule=rule, antithetic=None)
>>> a.shape
(1, 128)

>>> b = sampler.generator.generate_samples(order=128, domain=cpy_jac, rule=rule, antithetic=False)
>>> b.shape
(1, 65)

>>> c = cpy_jac.sample(size=128, rule=rule, antithetic=False, include_axis_dim=True)
>>> c.shape
(1, 65)

Expected behavior
There is no error. However, I can't wrap my head around why an identical call to generate_samples and sampler.generator.generate_samples results in different shapes!

Desktop (please complete the following information):

• OS: CentOS 7.9
• Python version: 3.8.8
• Chaospy version: 4.3.13

Additional context
See #420 (comment) for the pip freeze output on my system

[jonathf]

Thanks for the bug report.

This is indeed a bug in the antithetic variate module. I've made a fix in version 4.3.15. Please let me know if it solves the problem for you.