Scikit-Learn(sklearn)是 Python 科学计算与机器学习领域的核心开源工业级基石库,依托 NumPy、SciPy 与 Matplotlib 构建。其最核心的架构优势在于高度统一、严谨自洽的三大对象接口范式:
- Estimator(评估器):负责学习模型参数,核心方法为
fit(X, y)。 - Transformer(转换器):负责特征预处理与投影,核心方法为
transform(X)与fit_transform(X)。 - Predictor(预测器):负责在新样本上执行推断,核心方法为
predict(X)、predict_proba(X)与score(X, y)。
🔗 关联矩阵速查:NumPy 速查表 · Pandas 速查表 · 监督学习算法速查表 · 数据分析方法速查表 · Python 速查表
1. 端到端快速上手闭环示例
以下示例演示从加载内置数据集、切分训练/测试集、特征标准化、模型拟合、到预测与指标评估的完整闭环:
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score, classification_report
# 1. 加载数据
iris = load_iris()
X, y = iris.data, iris.target
# 2. 划分数据集 (分层抽样保持类别比例)
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42, stratify=y
)
# 3. 特征缩放 (严防数据泄漏:仅在训练集拟合)
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test)
# 4. 实例化评估器并拟合
knn = KNeighborsClassifier(n_neighbors=5)
knn.fit(X_train_scaled, y_train)
# 5. 推理预测与评估
y_pred = knn.predict(X_test_scaled)
print(f"准确率: {accuracy_score(y_test, y_pred):.4f}")
print(classification_report(y_test, y_pred, target_names=iris.target_names))
2. 数据准备与数据集划分
Scikit-Learn 接收规范的 2 维数组形状:特征矩阵 $X$ 形状为 (n_samples, n_features),目标标签 $y$ 形状为 (n_samples,)。支持 NumPy 数组、SciPy 稀疏矩阵(scipy.sparse)与 Pandas DataFrame / Series。
2.1 常见内置与生成数据集
from sklearn import datasets
# 内置小型基准数据集 (返回 Bunch 对象,类似字典)
iris = datasets.load_iris() # 鸢尾花分类
digits = datasets.load_digits() # 手写数字图像 (8x8)
diabetes = datasets.load_diabetes() # 糖尿病回归
wine = datasets.load_wine() # 葡萄酒多分类
# 现实大型数据集 (自动下载本地缓存)
housing = datasets.fetch_california_housing()
# 合成模拟数据 (算法调试与基准测试)
X, y = datasets.make_classification(
n_samples=1000, n_features=20, n_informative=15, n_classes=2, random_state=42
)
X_reg, y_reg = datasets.make_regression(n_samples=500, n_features=10, noise=0.1, random_state=42)
X_blob, y_blob = datasets.make_blobs(n_samples=300, centers=3, n_features=2, random_state=42)
2.2 数据集切分(train_test_split)
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(
X, y,
test_size=0.25, # 测试集比例 (默认 0.25)
random_state=42, # 固定随机种子确保结果可复现
shuffle=True, # 切分前打乱数据
stratify=y # 分类任务强烈建议启用:保持训练集与测试集标签分布一致
)
3. 特征工程与数据预处理(sklearn.preprocessing & sklearn.impute)
| 转换器类名 | 适用场景 | 核心数学转换 / 原理 | 关键参数 |
|---|---|---|---|
StandardScaler | 数值型特征 Z-score 标准化 | $z = \frac{x - \mu}{\sigma}$,转换为均值 0、方差 1 | with_mean=True, with_std=True |
MinMaxScaler | 特征限定在指定闭区间 | $x’ = \frac{x - x_{\min}}{x_{\max} - x_{\min}} \cdot (\max - \min) + \min$ | feature_range=(0, 1) |
RobustScaler | 含极端异常值的数值特征 | 基于四分位距缩放:$x’ = \frac{x - Q_2}{Q_3 - Q_1}$ | quantile_range=(25.0, 75.0) |
Normalizer | 样本行向量长度归一化 | 针对每个样本向量除以其 $L_1$、$L_2$ 或最大值范数 | norm='l2' ('l1', 'max') |
Binarizer | 连续值二值门限硬截断 | $x > \text{threshold} \Rightarrow 1$,否则 $0$ | threshold=0.0 |
SimpleImputer | 缺失值插补填补 | 均值、中位数、众数或固定常量填补 | missing_values=np.nan, strategy='mean' |
OneHotEncoder | 无序离散类别特征编码 | 将每个离散类别转为独立的二进制哑变量 (Dummy) 列 | drop='first', handle_unknown='ignore' |
OrdinalEncoder | 有序类别特征编码 | 将类别映射为连续整数 $0, 1, \dots, K-1$ | categories='auto', encoded_missing_value |
LabelEncoder | 目标变量 $y$ 的标签编码 | 将离散字符串标签编码为从 0 开始的整数 | 仅用于目标标签 $y$,切勿用于特征 $X$ |
PolynomialFeatures | 特征多项式与高阶交叉项 | 构造高阶特征 $[1, a, b, a^2, ab, b^2]$ 拟合非线性 | degree=2, include_bias=False |
3.1 预处理标准代码片段
import numpy as np
from sklearn.preprocessing import (
StandardScaler, MinMaxScaler, RobustScaler, Normalizer,
Binarizer, OneHotEncoder, OrdinalEncoder, LabelEncoder, PolynomialFeatures
)
from sklearn.impute import SimpleImputer
# 1. 数值缩放 (严禁在 test 上调用 fit)
scaler = StandardScaler()
X_train_std = scaler.fit_transform(X_train)
X_test_std = scaler.transform(X_test)
# 2. 缺失值填补 (采用 sklearn.impute.SimpleImputer)
imputer = SimpleImputer(missing_values=np.nan, strategy='median')
X_train_imp = imputer.fit_transform(X_train)
X_test_imp = imputer.transform(X_test)
# 3. 类别特征 One-Hot 编码 (处理测试集中未知类别的鲁棒配置)
ohe = OneHotEncoder(handle_unknown='ignore', sparse_output=False)
X_train_ohe = ohe.fit_transform(X_train_cat)
X_test_ohe = ohe.transform(X_test_cat)
# 4. 目标标签 y 的整型转换
le = LabelEncoder()
y_train_enc = le.fit_transform(y_train)
y_test_enc = le.transform(y_test)
# 5. 生成多项式特征
poly = PolynomialFeatures(degree=2, include_bias=False)
X_train_poly = poly.fit_transform(X_train)
4. 常用评估器速查(模型族分类)
4.1 监督学习:分类算法(Classification)
from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.naive_bayes import GaussianNB
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier, HistGradientBoostingClassifier
# 逻辑回归 (分类基线模型)
clf_lr = LogisticRegression(C=1.0, max_iter=200, random_state=42)
# K 近邻分类器
clf_knn = KNeighborsClassifier(n_neighbors=5, metric='minkowski', p=2)
# 支持向量机 (RBF / Linear 核)
clf_svm = SVC(C=1.0, kernel='rbf', gamma='scale', probability=True, random_state=42)
# 高斯朴素贝叶斯 (文本或小样本连续特征)
clf_gnb = GaussianNB()
# 随机森林 (Bagging 集成)
clf_rf = RandomForestClassifier(n_estimators=100, max_depth=None, n_jobs=-1, random_state=42)
# 基于直方图的梯度提升树 (GBDT 现代高效实现,对标 LightGBM)
clf_hgb = HistGradientBoostingClassifier(learning_rate=0.1, max_iter=100, random_state=42)
4.2 监督学习:回归算法(Regression)
from sklearn.linear_model import LinearRegression, Ridge, Lasso, ElasticNet
from sklearn.svm import SVR
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor, HistGradientBoostingRegressor
# 普通最小二乘线性回归
reg_lr = LinearRegression()
# L2 正则化岭回归 (防多重共线性)
reg_ridge = Ridge(alpha=1.0)
# L1 正则化 Lasso 回归 (特征稀疏化筛选)
reg_lasso = Lasso(alpha=0.1)
# 支持向量回归
reg_svr = SVR(kernel='rbf', C=1.0, epsilon=0.1)
# 随机森林回归器
reg_rf = RandomForestRegressor(n_estimators=100, n_jobs=-1, random_state=42)
4.3 无监督学习:聚类与降维(Clustering & Decomposition)
from sklearn.cluster import KMeans, DBSCAN
from sklearn.decomposition import PCA
# K-Means 聚类
kmeans = KMeans(n_clusters=3, init='k-means++', n_init=10, random_state=42)
kmeans.fit(X)
cluster_labels = kmeans.labels_ # 样本所属簇 ID
cluster_centers = kmeans.cluster_centers_ # 簇中心坐标
# 基于密度的空间聚类 (抗噪声,无需预设簇数)
dbscan = DBSCAN(eps=0.5, min_samples=5)
db_labels = dbscan.fit_predict(X) # -1 代表离群噪声点
# 主成分分析 (PCA) 降维
pca = PCA(n_components=0.95) # 保留 95% 累计方差贡献率
X_pca = pca.fit_transform(X)
print("保留主成分数量:", pca.n_components_)
print("各主成分解释方差比:", pca.explained_variance_ratio_)
5. 核心统一接口操作速查(API Cheat Sheet)
| 操作范式 | 典型方法调用 | 适用对象类型 | 说明与返回值 |
|---|---|---|---|
| 拟合状态 | estimator.fit(X, y) | 所有 Estimator | 在训练集上拟合并学习模型权重/统计量,返回自身 self |
| 推断预测 | predictor.predict(X) | 所有 Predictor | 输出硬标签(分类)或连续预测数值(回归),返回 NumPy 数组 |
| 概率评估 | predictor.predict_proba(X) | 分类器 (带概率标定) | 输出每个类别的后验概率分布矩阵,行和为 1 |
| 置信度量 | predictor.decision_function(X) | 线性模型 / SVM | 样本到分类超平面的带符号几何距离 |
| 特征变换 | transformer.transform(X) | 所有 Transformer | 基于已学得的参数转换输入特征矩阵并返回新特征矩阵 |
| 合并变换 | transformer.fit_transform(X) | 所有 Transformer | 等价于先 fit 后 transform(内部实现通常计算效率更高) |
| 默认评分 | estimator.score(X, y) | 所有分类/回归模型 | 分类器默认返回准确率(Accuracy);回归器默认返回拟合优度($R^2$) |
6. 模型评估指标体系(sklearn.metrics)
6.1 分类指标(Classification Metrics)
from sklearn.metrics import (
accuracy_score, precision_score, recall_score, f1_score,
roc_auc_score, roc_curve, confusion_matrix, classification_report,
ConfusionMatrixDisplay
)
# 1. 基础硬分类指标 (多分类需指定 average: 'binary', 'macro', 'micro', 'weighted')
acc = accuracy_score(y_test, y_pred)
prec = precision_score(y_test, y_pred, average='macro')
rec = recall_score(y_test, y_pred, average='macro')
f1 = f1_score(y_test, y_pred, average='macro')
# 2. 连续概率指标 (ROC-AUC 需要预测概率 y_prob)
y_prob = clf.predict_proba(X_test)[:, 1] # 二分类取正类概率
auc = roc_auc_score(y_test, y_prob)
# 3. 混淆矩阵与文本摘要报告
cm = confusion_matrix(y_test, y_pred)
report = classification_report(y_test, y_pred)
print(report)
# 4. 混淆矩阵可视化展示 (无需额外外部依赖)
disp = ConfusionMatrixDisplay(confusion_matrix=cm)
disp.plot()
6.2 回归指标(Regression Metrics)
from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score
mae = mean_absolute_error(y_test, y_pred)
mse = mean_squared_error(y_test, y_pred)
rmse = mean_squared_error(y_test, y_pred, squared=False) # 根均方误差 (RMSE)
r2 = r2_score(y_test, y_pred) # 判定系数 R²
6.3 聚类指标(Clustering Metrics)
from sklearn.metrics import (
adjusted_rand_score, homogeneity_score, v_measure_score, silhouette_score
)
# 外部指标 (存在真实参考标签 ground truth 时)
ari = adjusted_rand_score(y_true, cluster_labels) # 调整兰德指数 [-1, 1]
homo = homogeneity_score(y_true, cluster_labels) # 同质性 [0, 1]
vm = v_measure_score(y_true, cluster_labels) # V-measure [0, 1]
# 内部指标 (无真实标签,仅依赖几何距离)
sil = silhouette_score(X, cluster_labels) # 轮廓系数 [-1, 1]
7. 交叉验证与超参数寻优(sklearn.model_selection)
7.1 交叉验证评估(Cross-Validation)
from sklearn.model_selection import cross_val_score, cross_validate, StratifiedKFold
# 基础交叉验证得分
scores = cross_val_score(clf, X, y, cv=5, scoring='accuracy')
print(f"5折准确率: {scores.mean():.4f} (±{scores.std():.4f})")
# 自定义切分折叠器 (分类任务推荐分层切分)
skf = StratifiedKFold(n_splits=5, shuffle=True, random_state=42)
scores_skf = cross_val_score(clf, X, y, cv=skf, scoring='f1_macro')
# 多指标联合评估
multi_scores = cross_validate(
clf, X, y, cv=5,
scoring=['accuracy', 'roc_auc'],
return_train_score=False
)
print("平均测试 AUC:", multi_scores['test_roc_auc'].mean())
7.2 网格搜索与随机搜索(Hyperparameter Tuning)
from sklearn.model_selection import GridSearchCV, RandomizedSearchCV
from scipy.stats import randint, uniform
# 1. 网格搜索 (全量穷举搜索)
param_grid = {
'n_neighbors': [3, 5, 7, 11],
'weights': ['uniform', 'distance'],
'metric': ['euclidean', 'manhattan']
}
grid = GridSearchCV(
estimator=KNeighborsClassifier(),
param_grid=param_grid,
cv=5,
scoring='accuracy',
n_jobs=-1
)
grid.fit(X_train, y_train)
print("最佳参数:", grid.best_params_)
print("最佳得分:", grid.best_score_)
best_model = grid.best_estimator_
# 2. 随机参数搜索 (大参数空间高效抽样)
param_dist = {
'n_estimators': randint(50, 300),
'max_depth': [None, 5, 10, 20],
'min_samples_split': randint(2, 11)
}
rand_search = RandomizedSearchCV(
estimator=RandomForestClassifier(random_state=42),
param_distributions=param_dist,
n_iter=20, # 随机抽样参数组合次数
cv=5,
scoring='accuracy',
random_state=42,
n_jobs=-1
)
rand_search.fit(X_train, y_train)
print("随机搜索最佳参数:", rand_search.best_params_)
8. 工业实战标配:Pipeline 与 ColumnTransformer
8.1 为什么必须使用 Pipeline?
在机器学习工程实践中,若在切分数据集或交叉验证前对全局数据集执行了 fit_transform(如缺失值填补或标准化),验证集与测试集的信息会提前泄露至训练特征统计量中,造成模型离线评估虚高(数据泄漏 Data Leakage)。
Pipeline 能够将数据预处理步骤与模型估计器组装为一个原子化的评估器单元。在每一轮交叉验证折叠中,仅在当前折叠的训练集拟合转换器,并自动应用至当前验证折叠。
8.2 混合类型特征处理最佳模板
现实业务表格数据通常混杂了数值连续特征与离散类别特征,推荐采用 ColumnTransformer + Pipeline 组合拳方案:
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
# 1. 显式区分特征列名或列索引
numeric_features = ['age', 'income', 'credit_score']
categorical_features = ['gender', 'occupation', 'city']
# 2. 数值特征预处理子管线:填补中位数 -> 标准化缩放
numeric_transformer = Pipeline(steps=[
('imputer', SimpleImputer(strategy='median')),
('scaler', StandardScaler())
])
# 3. 类别特征预处理子管线:填补常数 -> 独热编码 (忽略未知类别)
categorical_transformer = Pipeline(steps=[
('imputer', SimpleImputer(strategy='constant', fill_value='missing')),
('onehot', OneHotEncoder(handle_unknown='ignore', sparse_output=False))
])
# 4. 列转换器整合:分别施加到对应列
preprocessor = ColumnTransformer(transformers=[
('num', numeric_transformer, numeric_features),
('cat', categorical_transformer, categorical_features)
])
# 5. 组装完整端到端模型管线
full_pipeline = Pipeline(steps=[
('preprocessor', preprocessor),
('classifier', RandomForestClassifier(n_estimators=100, random_state=42))
])
# 6. 一键拟合与预测 (防泄漏且推理极其优雅)
full_pipeline.fit(X_train_df, y_train)
y_pred = full_pipeline.predict(X_test_df)
print("Pipeline 准确率:", full_pipeline.score(X_test_df, y_test))