Wrong results with Scipy 1.13

[mancellin]

Describe the bug
Something changed between Scipy 1.12 and 1.13, that causes Chaospy to return clearly wrong values with the latest version of Scipy.

To Reproduce

$ pip install scipy==1.12
[...]
$ python -c 'import chaospy; print(chaospy.Normal().mom(2))'  
1.0000000000000002
$ pip install scipy==1.13
[...]
$ python -c 'import chaospy; print(chaospy.Normal().mom(2))'
[...]/lib/python3.12/site-packages/scipy/special/_basic.py:2930: RuntimeWarning: overflow encountered in scalar negative
  def corr(k, r): return np.power(k, -r / k) / gamma(r / k + 1) * r
[...]/lib/python3.12/site-packages/scipy/special/_basic.py:2930: RuntimeWarning: overflow encountered in power
  def corr(k, r): return np.power(k, -r / k) / gamma(r / k + 1) * r
inf

Expected behavior
The result returned by Scipy 1.12 is correct.

Desktop (please complete the following information):

• OS: Linux
• Python version: 3.12
• Chaospy version: 4.3.13

[omnivagant]

We noticed this in alpinelinux after upgrading our py3-scipy to 1.13.0, we consider downgrading to 1.12.1. There is an upcoming 1.13.1 release where it would be good if this was addressed, if it isn't already, https://github.com/scipy/scipy/milestone/84

[jonathf]

Looks like scipy.special.factorial2 can't handle uint32 dtypes anymore.

I've made a bug report here: scipy/scipy#20580

I see also some of the other tests I have set up fails. I'll have a look and see if we need to make some patches.

[jonathf]

Solved upstream.