Implement logcdf for TruncatedNormal

[LukeLB]

Closes #7003 first highlighted by a user on discord (https://discourse.pymc.io/t/censoring-a-truncated-distribution/13261) when trying to use the following model,

with pm.Model() as m:
    mu = pm.Normal("mu", 0, 1)
    sigma = pm.Normal("sigma", 0, 1)
    normal_ = pm.Normal.dist(mu, sigma)
    trunc_normal = pm.Truncated.dist(normal_, lower=0)
    censored = pm.Censored("censored", trunc_normal, lower=None, upper=upper_bounds, observed=cy)

By refactoring the truncated truncated_logcdf to take a base_rv_op we can now call it from TruncatedNormal. I've also moved the truncated_logcdf_from_base_rv out of truncated.py and into dist_math.py as I was having curcular import errors.

Checklist

• Explain important implementation details 👆
• Make sure that the pre-commit linting/style checks pass.
• Link relevant issues (preferably in nice commit messages)
• Are the changes covered by tests and docstrings?
• Fill out the short summary sections 👇

New features

-TruncatedNormal logcdf

---

📚 Documentation preview 📚: https://pymc--7034.org.readthedocs.build/en/7034/

[ricardoV94]

Thanks! Just suggesting a more direct test

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov]

Codecov Report

Merging #7034 (81dd599) into main (2e05854) will decrease coverage by 2.34%.
Report is 2 commits behind head on main.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7034      +/-   ##
==========================================
- Coverage   92.19%   89.86%   -2.34%     
==========================================
  Files         101      101              
  Lines       16893    16904      +11     
==========================================
- Hits        15575    15190     -385     
- Misses       1318     1714     +396     

Files	Coverage Δ
pymc/distributions/continuous.py	97.79% <100.00%> (+0.02%)	⬆️

... and 9 files with indirect coverage changes

[LukeLB]

I've rewritten the function and it returns the correct answers but not at the precision we would like, it is correct to 3 decimal places but past that the tests fail. Any idea why this might be occuring?

[ricardoV94]

I've rewritten the function and it returns the correct answers but not at the precision we would like, it is correct to 3 decimal places but past that the tests fail. Any idea why this might be occuring?

We are still probably testing too extreme values, did it fix the nan thing at least?

[LukeLB]

Yeah the nan problem is sorted, shall I just change the domains in the tests then? The domains in the check_logp tests seem to work fine, e.g.,

check_logp(
    pm.TruncatedNormal,
    R,
    {"mu": R, "sigma": Rplusbig, "lower": -Rplusbig, "upper": Rplusbig},
    scipy_logp,
    decimal=select_by_precision(float64=6, float32=1),
    skip_paramdomain_outside_edge_test=True,
        )

[ricardoV94]

Yup, sounds good

[ricardoV94]

Looks great

[LukeLB]

Any idea why these tests are failing? They all pass locally for me on my Mac

[ricardoV94]

It was a PyTensor issue that we already fixed, should work fine if we rerun them (I'm restarting them)