ColumnTransformer behavior for negative column indexes #12946
Description
Description
The behavior of ColumnTransformer when negative integers are passed as column indexes is not clear.
Steps/Code to Reproduce
import numpy as np
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder
X = np.random.randn(2, 2)
X_categories = np.array([[1], [2]])
X = np.concatenate([X, X_categories], axis=1)
print('---- With negative index ----')
ohe = OneHotEncoder(categories='auto')
tf_1 = ColumnTransformer([('ohe', ohe, [-1])], remainder='passthrough')
print(tf_1.fit_transform(X))
print('---- With positive index ----')
tf_2 = ColumnTransformer([('ohe', ohe, [2])], remainder='passthrough')
print(tf_2.fit_transform(X))Expected Results
The first transformer tf_1 should either raise an error or give the same result as the second transformer tf_2
Actual Results
---- With negative index ----
[[ 1. 0. 0.10600662 -0.46707426 1. ]
[ 0. 1. -1.33177629 2.29186299 2. ]]
---- With positive index ----
[[ 1. 0. 0.10600662 -0.46707426]
[ 0. 1. -1.33177629 2.29186299]]