cokelaer
Quality-of-Life Improvements: PEP 8 Standards and NumPy 2.0 Compatibility v2 plus I think some other fixes
Previously using histogram bin centers instead of the actual data to compute log-likelihood. That made AIC/BIC comparisons meaningless because likelihood was based on aggregated counts, not raw observations.
Fixed so it now runs dist.logpdf() on the real data array. AIC/BIC now reflect real statistical fit quality
The old code divided the standard deviation by sqrt(2π). That factor belongs to the Gaussian PDF normalization, not the variance itself.
Replaced with a proper weighted variance using bin frequencies as weights
Killed double-logging of log values in some branches. Cleaned up weighting logic so weighted stats are always coherent.
Please check these changes carefully, I might have made a mistake.
@tg12 looks great. Except the ruff stuff (see my comments) the rest looks good to me. would you mind remove it. I understand this is may be more powerful that pyproject, but I wish to keep the repo as simple as possible. Once the PR is accepted, I will update the README to add a contributor section where I will add your contrib.
Check you are satisfied with the changes. All tests pass. I want a second set of eyes on the math fixes. Ignore the ruff files if you want; they are only part of my development tooling. It is just a formatter and linter. If you are curious, it is here: https://github.com/astral-sh/ruff