← retour aux snippets

sklearn: Pipeline + ColumnTransformer

Prétraiter colonnes num/cat et entraîner un modèle.

python ml #sklearn#pipeline#preprocess

objectif

Prétraiter colonnes num/cat et entraîner un modèle.

code minimal

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.pipeline import Pipeline
from sklearn.linear_model import LogisticRegression
import numpy as np

X = np.array([[25,"FR"], [40,"DE"], [30,"FR"]], dtype=object)
y = np.array([0,1,0])
ct = ColumnTransformer([("num", StandardScaler(), [0]), ("cat", OneHotEncoder(handle_unknown="ignore"), [1])])
pipe = Pipeline([("prep", ct), ("clf", LogisticRegression(max_iter=1000))]).fit(X, y)
print(len(pipe.named_steps))

utilisation

print(pipe.predict([[35,"FR"]]).tolist())

variante(s) utile(s)

from sklearn.impute import SimpleImputer
ct2 = ColumnTransformer([("num", Pipeline([("imp", SimpleImputer()), ("sc", StandardScaler())]), [0])])
print(hasattr(ct2, "transform"))

notes

  • ColumnTransformer évite les DataFrames au profit de tableaux.