Shape issues when sampling prior predictive with MvNormal

[fbartolic]

I'm not sure if the this is intended or not, but I'm getting inconsistent shapes for a MvNormal RV depending on if I sample the prior with pm.sample or pm.sample_prior_predictive. Here's a minimal example:

import pymc3 as pm
import numpy as np

with pm.Model() as model:
    x = pm.MvNormal("x", mu=np.zeros(5), cov=np.eye(5), shape=(2, 5))
    trace = pm.sample(50)
    trace_pp = pm.sample_prior_predictive(50)
    
print(np.shape(trace['x'][0]))
print(np.shape(trace_pp['x'][0]))

I get the following output:

(2, 5)
(5,)

I would expect the shape of a sample from the predictive distribution to be (2, 5) rather than (5,).

Versions and main components

• PyMC3 Version: 3.8
• Theano Version: 1.0.4
• Python Version: 3.8
• Operating system: Mac
• How did you install PyMC3: pip

[lucianopaz]

Related to #2848 and #3706. The MvNormal.random method is buggy.

[canyon289]

Since this is related closing