CountEncoder bug when using Normalize and min_group_size

[mathisdrn]

I am not able to provide a simple reproducing test case but the following fail

from category_encoder import CountEncoder

# Works as expected
CountEncoder(
    drop_invariant=True,
    normalize=False,
    min_group_size=0.03,
    combine_min_nan_groups=True
).fit_transform(df[['a single column]])

# Works as expected
CountEncoder(
    drop_invariant=True,
    normalize=True,
    min_group_size=3,
    combine_min_nan_groups=True
).fit_transform(df[['a single column]])

# Doesn't work and returns an empty DataFrame with same number of rows
CountEncoder(
    drop_invariant=True,
    normalize=True,
    min_group_size=0.03,
    combine_min_nan_groups=True
).fit_transform(df[['a single column']])

weirdly enough I can't reproduce any of it with a simple test dataframe and i wasn't able to pinpoint any specific of my df column (dtype=object)

df = pd.DataFrame({
    'a': ['A', 'A', 'A', 'B', 'B', np.nan],
    'b': ['A', 'A', 'B', 'B', 'B', np.nan]
})

#3 works when switching drop_invariant from True to False but doesn't change the result of #2 and #1

• conclusion: drop_invariant have an effect on the result but only on normalize=True and min_group_size as percentage
  This does not make sense but i think it needs to be investigated.

I am resorting to not use normalize=True and instead place a standardscaler on the output.