Fix Stabilizing Predictions

[vmanc]

Pull Request

• [x] I confirm that I have read the CONTRIBUTING.md document.
• [x] I confirm that this contribution is made under an MIT license and that I have the authority necessary to make this contribution on behalf of its copyright owner.
• [x] I confirm that I did not use any AI-based coding assistance (such as, but not limited to, GitHub Copilot).

Description

[chschroeder]

Thank you very much for the PR!

A short summary of our e-mail correspondence, so that others can follow:

• This was part of the same e-mail correspondence as #51.
• The problem here was caused by me trying to fix the underlying problem: cohen's kappa can result in a division by zero when all predictions are of the same class (resulting in a nan value). My solution when I first tackled this issue, however, was not the correct way to handle this, and broke the original functionality.
• I offered to do the cleanup work (docstrings, changelog, and docs).
• Also I will add a new section to the docs (likely titled "errata") so that this bug is documented.
• We agreed that I can move adapted_cohen_kappa() to the outer level so that it could be imported from other places as well.