Please consider releasing kuma_utils as a PyPI package.

Hello! Would you accept a PR? I use LGBMImputer on large datasets (>>1000 columns) and I'd like to wrap the loop over columns in tqdm.

In my project, I have implemented the LGBMImputer in your package for data imputation. And as usual, I first fit and transform the train set, then transform the test set.

ISSUE: The columns with missing data in my train set and in test set are different and it appears that the "fit" function only train model and regress on the feature_with_missing of train data, and left several other columns in test set with data still missing if those columns is not in the feature_with missing from train set.

The solution I have applied in my project is to 'fit' regression on all train columns, and 'transform' feature missing (see attachment for my changed codes).
I'm not yet adept in coding, still, I'd imagine it may increase the execution time so I will leave this issue here for any better solution.

imputer.txt

I used your module for a project and it returned an error : "Truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all()".

The error comes from the line : self.feature_with_missing = [ col for col in self.feature_names if X[col].isnull().sum() > 0]

Replacing it with :

nan_count = X.isna().sum()
self.feature_with_missing = nan_count[nan_count > 0].index.tolist()

solves the issue