Styjun

I'm using target encoding on some features in my dataset. My full pipeline is as such:

from sklearn.compose import ColumnTransformer

from sklearn.pipeline import Pipeline

from sklearn.preprocessing import OneHotEncoder
from sklearn.preprocessing import StandardScaler

from category_encoders.target_encoder import TargetEncoder

from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import train_test_split

numeric_features = ['feature_1']
numeric_pipeline = Pipeline(steps=[('scaler', StandardScaler())])

ohe_features = ['feature_2', 'feature_3', 'feature_4']
ohe_pipeline = Pipeline(steps=[('ohe', OneHotEncoder())])

te_features = ['feature_5', 'feature_6']
te_pipeline = TargetEncoder()

preprocessor = ColumnTransformer(transformers=[
 ('numeric', numeric_pipeline, numeric_features), 
 ('ohe_features', ohe_pipeline, ohe_features), 
 ('te_features', te_pipeline, te_features)
 ]
 )

clf_lr = Pipeline(steps=[
 ('preprocessor', preprocessor), 
 ('classifier', LogisticRegression())
 ]
 )

X_train, X_test, y_train, y_test = train_test_split(df_testing.drop(columns='target'), 
 df_testing['target'], 
 stratify=df_testing['target'])

params = 'classifier__C': [0.001, 0.01, 0.05, 0.1, 1]

gs = GridSearchCV(clf_lr, params, cv=3)
gs.fit(X_train, y_train)

The problem is that the call to the fit method in GridSearchCV is failing because of the TargetEncoder step in the pipeline. Specifically, it is throwing

IndexingError: Unalignable boolean Series provided as indexer (index of the boolean Series and of the indexed object do not match

Even when I call reset_index(drop=True) on both X_train and y_train, I get this error.

If I just call:

clf_lr.fit(X_train.reset_index(drop=True), y_train.reset_index(drop=True))
clf_lr.score(X_test.reset_index(drop=True), y_train.reset_index(drop=True)) # both calls to reset_index required otherwise the same IndexingError is thrown

the code works. However, I need the cross validation to find the best parameter C for LogisticRegression. The same would apply for cross validation on any other model I wish to try.

Could anyone please let me know if this is a known issue with TargetEncoder or if I've implemented or fitted my pipeline incorrectly?