Fix size issue when using broadcastable dimensions in multivariate RandomVariables

[ricardoV94]

Closes #1055

The original issue seems to have been misunderstood. This PR allows size to broadcast multivariate distribution parameters in exactly the same way that it broadcasts univariate distribution parameters. It does not allow size to be shorter or inconsistent with the original parameter shapes, which seems to be the reason why the more restrictive checks were introduced originally. The following cases are now both possible:

import numpy as np
import aesara.tensor as at

# Already allowed before
at.random.normal(np.ones((3, 1, 3)), size=(3, 10, 3)).eval().shape  # (3, 10, 3)

# Allowed now
at.random.dirichlet(np.ones((3, 1, 3)), size=(3, 10)).eval().shape  # (3, 10, 3)

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79.25%. Comparing base (be222f0) to head (c3a34ea).
Report is 503 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1056      +/-   ##
==========================================
- Coverage   79.25%   79.25%   -0.01%     
==========================================
  Files         152      152              
  Lines       47882    47875       -7     
  Branches    10909    10906       -3     
==========================================
- Hits        37949    37942       -7     
  Misses       7436     7436              
  Partials     2497     2497              

Files	Coverage Δ
aesara/tensor/random/basic.py	98.90% <100.00%> (-0.02%)	⬇️

[brandonwillard]

The original issue seems to have been misunderstood. This PR allows size to broadcast multivariate distribution parameters in exactly the same way that it broadcasts univariate distribution parameters. It does not allow size to be shorter or inconsistent with the original parameter shapes, which seems to be the reason why the more restrictive checks were introduced originally. The following cases are now both possible:

No, the original issue was not misunderstood. See #1044 (reply in thread). I'm now assuming you mean the first issue you created that was converted into a discussion, and not the issue(s) addressed by #446.

I've clarified the issue in #1058 and separated it from our general interpretation of size (i.e. #412). That interpretation shouldn't be an issue; instead, there only seems to be a bug in our current implementation.

[ricardoV94]

Oh, I forgot to fix the Categorical distribution as well