Incorrect shape handling in `scipy.stat.multivariate_t.rvs` method

The rvs method of multivariate_t works only when size parameter is strictly 1 dimensional. This is not the case with multivariate_normal.

Reproducing code example:

>>> import sys, scipy, numpy; print(scipy.__version__, numpy.__version__, sys.version_info)
1.6.3 1.20.3 sys.version_info(major=3, minor=7, micro=7, releaselevel='final', serial=0)
>>> import numpy as np
>>> from scipy.stats import multivariate_normal, multivariate_t
>>> mean, cov = np.zeros(2), np.eye(2)
>>> multivariate_normal.rvs(mean=mean, cov=cov, size=None).shape
(2,)
>>> multivariate_t.rvs(loc=mean, shape=cov, size=None)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/sayam/miniconda3/envs/pymc_dev/lib/python3.7/site-packages/scipy/stats/_multivariate.py", line 4148, in rvs
    samples = loc + z / np.sqrt(x)[:, None]
IndexError: invalid index to scalar variable.

>>> multivariate_normal.rvs(mean=mean, cov=cov, size=(10, 3)).shape
(10, 3, 2)
>>> multivariate_t.rvs(loc=mean, shape=cov, size=(10, 3))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/sayam/miniconda3/envs/pymc_dev/lib/python3.7/site-packages/scipy/stats/_multivariate.py", line 4148, in rvs
    samples = loc + z / np.sqrt(x)[:, None]
ValueError: operands could not be broadcast together with shapes (10,3,2) (10,1,3) 

>>> multivariate_t.rvs(loc=mean, shape=cov, size=(3))
array([[ 1.54502738, -0.23866557],
       [-0.79040924, -1.72035808],
       [-0.28060545, -0.69889876]])

I guess the errors arise from this line - https://github.com/scipy/scipy/blob/f856e3e852a59120d44133a0ce550cceb19f9244/scipy/stats/_multivariate.py#L4084

This issue popped up while adding MvStudentT to PyMC3 https://github.com/pymc-devs/pymc3/pull/4731. Its random sampling method makes use of scipy.stats.multivariate_t.

[Sayam753]

Unfortunately, I am on MacOS and was following this tutorial.

• In step 1, I installed miniconda instead of Anaconda (trying to be minimalist here)
• In step 11, this command conda install mkl openblas cython numpy pytest pybind11 pythran=0.9.11 -c conda-forge helped me to follow next steps.

Sorry, I was not aware you were working on this.

My bad, I should have mentioned this earlier. Thanks for the helping hands anyways.

[tirthasheshpatel]

I was also thinking about the use of ellipsis to fix this issue but struggled my way through setting up scipy dev environment locally. FInally, I got it working.

Happy to hear that you got it working! You can refer to https://docs.scipy.org/doc/scipy/reference/building/linux.html#debian-ubuntu anytime if you are on Ubuntu/Debian. If you are on windows, this wiki page might help set it up easily. Sorry, I was not aware you were working on this.

Now, I’ll try to understand the shape handling in scipy in near future crossed_fingers

Sure, @Sayam753 😃! It would be very helpful if you could resolve some of the shape bugs I mentioned above. If you face any problems, feel free to ping me.