Features not transformed by BinningProcess

[jnsofini]

It will be great to have a way to have special features that are not transformed by the binning process. This gives users a way to have columns like snapshot-date and ids (not used in calculating woe) that are not transformed but can be easily mapped to the raw data for use in the downstream process.

[guillermo-navas-palencia]

Hi @jnsofini.

I think users can perform the operation you described leveraging sklearn.compose.ColumnTransformer. See the example below. Given the versatility provided by ColumnTransformer, I am not entirely sure whether it is worth incorporating a similar function. Happy to discuss this further.

import numpy as np
import pandas as pd

from optbinning import BinningProcess
from sklearn.compose import ColumnTransformer
from sklearn.datasets import load_breast_cancer

data = load_breast_cancer()
df = pd.DataFrame(data.data, columns=data.feature_names)
y = data.target

Add id column

df['id'] = np.arange(df.shape[0])

Use column transformer and pass the features to apply binning. The remaining column (id) will be passed through.

features = df.columns[:-1].tolist()

binning_process = BinningProcess(variable_names=features)

ct = ColumnTransformer(
    transformers=[('binning_process', binning_process, features)],
    remainder='passthrough'
)

Xt = ct.fit_transform(df, y)

Convert to dataframe

pd.DataFrame(data=Xt, columns=df.columns)

[guillermo-navas-palencia]

These links are relevant. I see the value of this proposed feature:

• https://stackoverflow.com/questions/74020622/python-optbinning-packages-optimalbinning-and-binningprocess-giving-different-r/74037157#74037157
• https://stackoverflow.com/questions/68874492/preserve-column-order-after-applying-sklearn-compose-columntransformer