BUG: Combination covariance errors with WhiteNoise

[dehorsley]

Describe the issue:

Post v5.2.0, gp.cov.Combination covariance has extra checks (at cov.py:189) that fail with a WhiteNoise covariance.

Currently, the check for input_dim equality fails there is also a lurking problem with calculating active_dims.

Reproduceable code example:

import pm
pm.gp.cov.Exponential(2, ls=1) + pm.gp.cov.WhiteNoise(sigma=1)

Error message:

AttributeError                            Traceback (most recent call last)

<ipython-input-1-7257c4d70473> in <cell line: 3>()
      1 import pymc as pm
      2 print(pm.__version__)
----> 3 pm.gp.cov.Exponential(2, ls=1) + pm.gp.cov.WhiteNoise(sigma=1)

2 frames

/usr/local/lib/python3.9/dist-packages/pymc/gp/cov.py in __add__(self, other)
     83         if isinstance(other, numbers.Real):
     84             other = Constant(c=other)
---> 85         return Add([self, other])
     86 
     87     def __mul__(self, other):

/usr/local/lib/python3.9/dist-packages/pymc/gp/cov.py in __init__(self, factor_list)
    187 
    188         # Check if all input_dim are the same in factor_list
--> 189         input_dims = {factor.input_dim for factor in factor_list if isinstance(factor, Covariance)}
    190 
    191         if len(input_dims) != 1:

/usr/local/lib/python3.9/dist-packages/pymc/gp/cov.py in <setcomp>(.0)
    187 
    188         # Check if all input_dim are the same in factor_list
--> 189         input_dims = {factor.input_dim for factor in factor_list if isinstance(factor, Covariance)}
    190 
    191         if len(input_dims) != 1:

AttributeError: 'WhiteNoise' object has no attribute 'input_dim'

PyMC version information:

pymc==5.2.0

installed via pip

OS: *

Context for the issue:

Affects any use of latent GPs with noise.

[dehorsley]

Likely also an issue with Constant covariance kernel

Edit: actually no, because Constant is a subclass of BaseCovariance, not Covariance. It looks like it was intended for WhiteNoise to do the same.

[ricardoV94]

Closed via #6674