Machine Learning
ColumnTransformer
ColumnTransformer applies different transforms to different columns.
ColumnTransformer
In real tables columns need different processing: numbers are scaled, categories one-hot encoded. ColumnTransformer applies a separate transformer to each column group and concatenates the results side by side:
from sklearn.compose import ColumnTransformer
ct = ColumnTransformer([
("sayi", StandardScaler(), ["yas", "gelir"]),
("kat", OneHotEncoder(handle_unknown="ignore"), ["sehir"]),
])
Z = ct.fit_transform(df) # 2 scaled + as many 0/1 columns as cities
Columns can be given by pandas names or integer positions.
A full preprocessing chain
Each column group can have its own sub-pipeline; the whole structure is gathered with the model into a single pipeline:
sayisal = make_pipeline(SimpleImputer(strategy="median"), StandardScaler())
kategorik = make_pipeline(SimpleImputer(strategy="most_frequent"),
OneHotEncoder(handle_unknown="ignore"))
ct = ColumnTransformer([("sayi", sayisal, sayi_sutunlar),
("kat", kategorik, kat_sutunlar)])
model = make_pipeline(ct, LogisticRegression())
get_feature_names_out() shows what each column represents after the transformation.
Exercise Transform a mixed table
Build a ColumnTransformer processing the yas and gelir columns with StandardScaler and the sehir column with OneHotEncoder. Print the shape of the fit_transform result, then the get_feature_names_out() list.
Show example solution
import numpy as np
import pandas as pd
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import StandardScaler, OneHotEncoder
df = pd.DataFrame({
"yas": [25, 32, 47, 51],
"gelir": [3000, 4500, 8000, 6000],
"sehir": ["Ankara", "İzmir", "Ankara", "Bursa"],
})
ct = ColumnTransformer([("sayi", StandardScaler(), ["yas", "gelir"]), ("kat", OneHotEncoder(), ["sehir"])])
Z = ct.fit_transform(df)
print(Z.shape)
print(list(ct.get_feature_names_out()))
Test yourself
-
What does ColumnTransformer do by default to columns not named in the list?
- Scales them
- Errors
- Drops them (remainder="drop")
- Passes them through unchanged
Show answer
Drops them (remainder="drop") — They can be kept with remainder="passthrough".
-
Keep the columns that aren't named.
ct = ColumnTransformer([...], remainder=____)
Show answer
ct = ColumnTransformer([...], remainder="passthrough")
-
Match the parts.
- ("sayi", StandardScaler(), [...])
- Numeric transformation
- ("kat", OneHotEncoder(), [...])
- Categorical transformation
- remainder
- Remaining columns
- get_feature_names_out
- New column names
Next lesson: Summary - Pipeline
