Add user-friendly logp and logcdf methods

[ricardoV94]

This PR adds a logp helper similar to the logcdf that already existed to facilitate the extraction of logp and logcdf expressions, similar to how the .logp and .logcdf methods worked in v3:

import pymc3 as pm

# v3:
logp_norm = pm.Normal.dist(0, 1).logp(5.0)
locdf_norm = pm.Normal.dist(0, 1).logcdf(5.0)

# v4:
logp_norm = pm.logp(pm.Normal.dist(0, 1), 5.0)
logcdf_norm = pm.logcdf(pm.Normal.dist(0, 1), 5.0)

The file distributions/logp.py was renamed to distributions/logprob.py to avoid the the name conflict.

[michaelosthege]

I like it!

That reduces one breaking v3→v4 change to something less painful:

my_rv = pm.Normal.dist(0, 2)

# on v3:
my_rv.logp(1.5).eval()

# on v4.
pm.logp(my_rv, 1.5).eval()

[brandonwillard]

Generally speaking, don't do too much with respect to the likelihood calculations in PyMC, because we're ultimately going to set PyMC up to use aeppl. Since this looks more like a PyMC-specific interface change, that might not be an issue, though.

[ricardoV94]

Generally speaking, don't do too much with respect to the likelihood calculations in PyMC, because we're ultimately going to set PyMC up to use aeppl. Since this looks more like a PyMC-specific interface change, that might not be an issue, though.

Yeah, this one is actually mimicking a bit what we have in aeppl just to extract a logp/logcdf of a vanilla RV

[codecov]

Codecov Report

Merging #4833 (528edc7) into main (223b0b8) will increase coverage by 0.01%.
The diff coverage is 90.90%.

@@            Coverage Diff             @@
##             main    #4833      +/-   ##
==========================================
+ Coverage   72.31%   72.32%   +0.01%     
==========================================
  Files          85       85              
  Lines       13877    13884       +7     
==========================================
+ Hits        10035    10042       +7     
  Misses       3842     3842              

Impacted Files	Coverage Δ
pymc3/distributions/logprob.py	95.89% <88.88%> (ø)
pymc3/distributions/__init__.py	100.00% <100.00%> (ø)
pymc3/distributions/discrete.py	99.00% <100.00%> (ø)
pymc3/step_methods/hmc/base_hmc.py	91.05% <0.00%> (+0.81%)	⬆️

[michaelosthege]

bad luck