truncnorm gives incorrect means and variances

truncnorm.mean and truncnorm.var appear to give incorrect outputs when loc or scale are array-like and contain more than one number:

>>> import scipy.stats as stats
>>> stats.truncnorm.mean(-1, 2, loc=0)
0.229637179091329
>>> stats.truncnorm.mean(-1, 2, loc=[0])
array([0.22963718])
>>> stats.truncnorm.mean(-1, 2, loc=[0, 0]) # expect to get 0.2296 twice
array([0., 0.])
>>> stats.truncnorm.var(-1, 1, loc=0)
0.29112509477279314
>>> stats.truncnorm.var(-1, 1, loc=[0])
array([0.29112509])
>>> stats.truncnorm.var(-1, 1, loc=[0, 0]) # expect to get 0.2911 twice
array([1., 1.])

truncnorm.moment doesn’t seem to support array-likes at all, either:

>>> stats.truncnorm.moment(1, -1, 1, loc=[0])
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\ProgramData\Miniconda3\lib\site-packages\scipy\stats\_distn_infrastructure.py", line 1266, in moment
    fac = float(scale) / float(loc)
TypeError: float() argument must be a string or a number, not 'list'

This is on Scipy version 1.6.0. This problem was even worse on version 1.4.1, where truncnorm.var() could produce negative numbers:

>>> stats.truncnorm.var(-1, 1, loc=0)
0.29112509477279314
>>> stats.truncnorm.var(-1, 1, loc=[0, 0])
array([-0.41774981, -0.41774981])

I don’t think I’m using these functions incorrectly, as the docstrings for mean(), var(), and moment() all say that loc and scale can be array-like.

[mdhaber]

One problem is line 7917:

        pA, pB = self._pdf(np.array([a, b]), a, b)

fix for mean is:

        pA, pB = self.pdf(np.array([a, b]), a, b)

But that didn’t fix var so I’ll just use np.vectorize. I’ll open a PR.

[czgdp1807]

Ah! I missed it. No problems. Thanks for highlighting your work on this issue.