python机器学习 Pipeline

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression

# 数据
X, y = load_iris(return_X_y=True)

# 切训练 / 测试
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

# Pipeline
pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('model', LogisticRegression(max_iter=200))
])

# 训练
pipe.fit(X_train, y_train) # X_train → scaler.fit_transform → model.fit

# 预测
print(pipe.predict(X_test)) # X_test → scaler.transform → model.predict
print(pipe.score(X_test, y_test))

from sklearn.feature_selection import SelectKBest, f_classif

pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('select', SelectKBest(f_classif, k=2)),
    ('model', LogisticRegression())
])

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder

preprocess = ColumnTransformer([
    ('num', StandardScaler(), ['age', 'income']),
    ('cat', OneHotEncoder(), ['gender'])
])

pipe = Pipeline([
    ('preprocess', preprocess),
    ('model', LogisticRegression())
])

from sklearn.pipeline import make_pipeline
# 标准语法
pipe_long = Pipeline([("scaler", MinMaxScaler()), ("svm", SVC(C=100))])
# 缩写语法
pipe_short = make_pipeline(MinMaxScaler(), SVC(C=100))

pipe = make_pipeline(StandardScaler(), LogisticRegression())
param_grid = {'logisticregression__C': [0.01, 0.1, 1, 10, 100]}
X_train, X_test, y_train, y_test = train_test_split(
    cancer.data, cancer.target, random_state=4)
grid = GridSearchCV(pipe, param_grid, cv=5)
grid.fit(X_train, y_train)
print("Best estimator:\n{}".format(grid.best_estimator_))