Logprob derivation for switch-encoding graphs

[shreyas3156]

What is this PR about?
This draft PR consists of the headway made in inferring the log-probability of switch-encoding expressions for the following cases:

1. when both branches have an encoding constant,

x = pt.random.normal(0.3)
y = pt.switch(x > 0.5, 1, 2)

2. when one of the two branches is measurable.

x = pt.random.normal(0.3)
y = pt.switch(x > 0.5, 1, x)

The pending challenges with the second case above include the following:

1. Add calculation for the logprob when one of the branches is measurable,
2. Get the pt.invert to go through the IR rewrite and convert to its Measurable counterpart.

@ricardoV94 @larryshamalama

[codecov]

Codecov Report

Merging #6834 (dc50061) into main (14e673f) will decrease coverage by 4.64%.
Report is 59 commits behind head on main.
The diff coverage is 27.63%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6834      +/-   ##
==========================================
- Coverage   91.89%   87.26%   -4.64%     
==========================================
  Files          95       96       +1     
  Lines       16185    16518     +333     
==========================================
- Hits        14874    14415     -459     
- Misses       1311     2103     +792     

Files Changed	Coverage Δ
pymc/logprob/censoring.py	44.65% <27.63%> (-54.23%)	⬇️

... and 36 files with indirect coverage changes

[larryshamalama]

Great start @shreyas3156 :) Some minor comments for now as you continue to make progress. Keep up the good work!

[ricardoV94]

Hey @shreyas3156 what's the status of the PR? Ready for review? There's some conflicts with main

[shreyas3156]

Hey @shreyas3156 what's the status of the PR? Ready for review? There's some conflicts with main

This was the first draft PR which was supposed to be a draft for the main PR #6949 , which is ready for review. We can close this one.