Handle falsy explicit impute replacements

[driavysinus]

Problem

explicit_impute() accepts mapping replacement values typed as str | int | float, but falsy values such as 0 are skipped because the current implementation checks replacement values by truthiness.

For example, replacement={'intcol': 0} should be a valid per-column replacement, but 0 is treated as if no replacement was configured.

Fix

Use mapping key existence checks in _extract_impute_value() and is not None in the caller, so valid falsy replacements are applied. Added a regression test for replacement={'intcol': 0}.

Validation

.venv\Scripts\pytest.exe tests\preprocessing\test_imputation.py::test_explicit_impute_subset_accepts_falsy_replacement_value -q
# 2 passed

.venv\Scripts\pytest.exe tests\preprocessing\test_imputation.py::test_explicit_impute_subset -q
# 2 passed

python -m py_compile ehrapy\preprocessing\_imputation.py tests\preprocessing\test_imputation.py

Risk

Low. This changes behavior only for mapping replacement values that are valid by type but were previously skipped because they are falsy.

[driavysinus]

The GPU tests pass check did not execute the GPU suite; it failed at the fork-permission gate in .github/workflows/test_gpu.yml. The maintainer-applied skip-gpu-ci label reflects that.

[sueoglu]

Thanks!

[sueoglu]

It would be nice to test other falsy values than only 0, what do you think?

[eroell]

I see the point. Thank you!

Additional regression test is great, with all previous tests still passing: as this is of course correcting the function towards intended behavior, although for quite a special case.

Please add a release note - this is also to give your work visible credit!

[eroell]

Seems here on the CI some tests fail, which are most likely unrelated and are due to new releases.

For miceforest, the problem is with the upstream package itself:

AnotherSamWilson/miceforest#104