Scikit-learn：一套 fit / predict 走天下

from sklearn.datasets import load_wine
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier

X, y = load_wine(return_X_y=True)     # 数据:直接拿到 numpy 数组

X_tr, X_te, y_tr, y_te = train_test_split(
    X, y, test_size=0.25, random_state=0, stratify=y)

clf = RandomForestClassifier(random_state=0)
clf.fit(X_tr, y_tr)                   # 训练:一行
print(clf.score(X_te, y_te))          # 评分:还是一行,这份数据能到 1.0
print(clf.predict(X_te[:3]))          # 预测新样本的类别

from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import SVC

for model in [LogisticRegression(max_iter=1000),
              RandomForestClassifier(random_state=0),
              SVC()]:
    model.fit(X_tr, y_tr)              # 同一个 fit
    print(model.__class__.__name__,
          model.score(X_te, y_te))     # 同一个 score

from sklearn import datasets

iris   = datasets.load_iris()          # 150x4 鸢尾花三分类
digits = datasets.load_digits()        # 1797 个 8x8 手写数字
wine   = datasets.load_wine()          # 178x13 红酒三分类

# 回归练习用加州房价:20640 条 x 8 特征,首次运行联网下载
housing = datasets.fetch_california_housing()

# 旧教程里的 load_boston 已在 1.2 版被移除(伦理问题),别再用
Xs, ys = datasets.make_classification(n_samples=1000, random_state=0)

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.linear_model import LogisticRegression

pipe = Pipeline([
    ("scaler", StandardScaler()),        # 第一步:标准化
    ("model",  LogisticRegression()),    # 第二步:模型
])

pipe.fit(X_tr, y_tr)      # 一次调用:先 fit 缩放,再训练模型
pipe.score(X_te, y_te)    # 评分时同样的顺序自动重放

from sklearn.model_selection import cross_val_score, GridSearchCV

print(cross_val_score(pipe, X_tr, y_tr, cv=5).mean().round(3))

param_grid = {"model__C": [0.1, 1, 10]}
#          model 这一步的 C 参数 —— 双下划线 = 「走进去」

grid = GridSearchCV(pipe, param_grid, cv=5, n_jobs=-1)
grid.fit(X_tr, y_tr)
print(grid.best_params_, grid.best_score_.round(3))

import numpy as np
import pandas as pd
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split, cross_val_score, GridSearchCV

rng = np.random.default_rng(42)
n = 600
df = pd.DataFrame({
    "age":    rng.integers(18, 70, n).astype(float),
    "income": rng.normal(6000, 2000, n).round(2),
    "city":   rng.choice(["北京", "上海", "广州", "深圳"], n),
    "device": rng.choice(["ios", "android", "web"], n),
})
df.loc[df.sample(frac=0.10, random_state=0).index, "device"] = np.nan
df.loc[df.sample(frac=0.05, random_state=1).index, "income"] = np.nan

# 目标:与特征相关的规则 + 噪声,模拟真实标签
filled = df["income"].fillna(df["income"].median())
logit = (filled - 6000) / 1500 + (df["city"] == "上海") * 0.8
y = (logit + rng.normal(0, 1, n) > 0.8).astype(int)

num_cols = ["age", "income"]
cat_cols = ["city", "device"]

preprocess = ColumnTransformer(transformers=[
    ("num", Pipeline([
        ("impute", SimpleImputer(strategy="median")),
        ("scale",  StandardScaler())]), num_cols),
    ("cat", Pipeline([
        ("impute", SimpleImputer(strategy="most_frequent")),
        ("onehot", OneHotEncoder(handle_unknown="ignore"))]), cat_cols),
])

clf = Pipeline([("prep", preprocess),
                ("model", RandomForestClassifier(random_state=0))])

X_tr, X_te, y_tr, y_te = train_test_split(
    df, y, test_size=0.2, random_state=42, stratify=y)

# 交叉验证:整条管道(含填充、缩放、编码)每折重新 fit,无泄漏
cv = cross_val_score(clf, X_tr, y_tr, cv=5)
print(cv.mean().round(3), cv.std().round(3))

# 网格调参:双下划线直达管道深处的模型参数
param_grid = {"model__n_estimators": [200, 500],
              "model__max_depth": [None, 8]}
search = GridSearchCV(clf, param_grid, cv=5, n_jobs=-1)
search.fit(X_tr, y_tr)
print(search.best_params_)

# 测试集终评(只做一次);顺手看 onehot 展开后的特征名
print(round(search.score(X_te, y_te), 3))
names = search.best_estimator_.named_steps["prep"].get_feature_names_out()
print(list(names)[:6])

# 错:全量数据先标准化,测试集信息已泄漏
X_scaled = StandardScaler().fit_transform(X)

# 对:交给管道,分折处理
pipe = Pipeline([("s", StandardScaler()), ("m", SVC())])
cross_val_score(pipe, X, y, cv=5)

# 不分层:小数据集可能切出「某类样本不足」的测试集
X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.25)

# 分层:两份数据的类别比例与全量一致,分类任务的默认写法
X_tr, X_te, y_tr, y_te = train_test_split(
    X, y, test_size=0.25, stratify=y)

joblib.dump(clf, "model.joblib")
# 加载端 sklearn 版本不一致时,可能直接报错或行为变化
python -c "import sklearn; print(sklearn.__version__)"