# 机器学习入门：Scikit-learn实战指南

> 作者：机器学习实战
> 标签：#Python #机器学习 #Scikit-learn #实战教程  
> 阅读时间：12分钟  
> 难度：⭐⭐⭐⭐ (适合有Python基础)

## 前言

机器学习不再是遥不可及的技术。Scikit-learn让Python开发者能快速构建ML模型。

本文通过实战案例，带你从零开始掌握Scikit-learn核心技能！

---

## 1. Scikit-learn简介

### 1.1 什么是Scikit-learn？

Scikit-learn（sklearn）是Python最流行的机器学习库，提供：
- 监督学习：分类、回归
- 无监督学习：聚类、降维
- 模型选择与评估
- 数据预处理

### 1.2 安装

```bash
pip install scikit-learn
```

---

## 2. 数据预处理

### 2.1 标准化

```python
from sklearn.preprocessing import StandardScaler
import numpy as np

# 创建数据
X = np.array([[1, -1, 2], 
              [2, 0, 0], 
              [0, 1, -1]])

# 标准化（均值0，方差1）
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

print("原始数据：\n", X)
print("标准化后：\n", X_scaled)
```

### 2.2 归一化

```python
from sklearn.preprocessing import MinMaxScaler

# 归一化到[0, 1]
scaler = MinMaxScaler()
X_normalized = scaler.fit_transform(X)
```

### 2.3 编码分类变量

```python
from sklearn.preprocessing import LabelEncoder

# 标签编码
le = LabelEncoder()
y = le.fit_transform(['cat', 'dog', 'cat', 'dog', 'bird'])
print(y)  # [1 2 1 2 0]

# 独热编码
from sklearn.preprocessing import OneHotEncoder
ohe = OneHotEncoder()
Y = ohe.fit_transform(y.reshape(-1, 1)).toarray()
```

---

## 3. 监督学习：分类

### 3.1 鸢尾花分类（经典案例）

```python
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
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. 划分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.3, random_state=42
)

# 3. 创建模型
knn = KNeighborsClassifier(n_neighbors=3)

# 4. 训练模型
knn.fit(X_train, y_train)

# 5. 预测
y_pred = knn.predict(X_test)

# 6. 评估
print("准确率：", accuracy_score(y_test, y_pred))
print("\n分类报告：\n", classification_report(y_test, y_pred))
```

### 3.2 支持向量机（SVM）

```python
from sklearn.svm import SVC

# 创建SVM分类器
svm = SVC(kernel='rbf', C=1.0, gamma='auto')

# 训练
svm.fit(X_train, y_train)

# 预测
y_pred = svm.predict(X_test)

# 评估
print("SVM准确率：", accuracy_score(y_test, y_pred))
```

### 3.3 随机森林

```python
from sklearn.ensemble import RandomForestClassifier

# 创建随机森林
rf = RandomForestClassifier(
    n_estimators=100,  # 树的数量
    max_depth=5,       # 最大深度
    random_state=42
)

# 训练
rf.fit(X_train, y_train)

# 特征重要性
print("特征重要性：", rf.feature_importances_)

# 预测
y_pred = rf.predict(X_test)
```

---

## 4. 监督学习：回归

### 4.1 线性回归

```python
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_squared_error, r2_score
import numpy as np

# 创建示例数据
X = np.array([[1], [2], [3], [4], [5]])
y = np.array([2, 4, 6, 8, 10])

# 创建模型
model = LinearRegression()

# 训练
model.fit(X, y)

# 预测
y_pred = model.predict(X)

# 评估
print("系数：", model.coef_)
print("截距：", model.intercept_)
print("MSE：", mean_squared_error(y, y_pred))
print("R²：", r2_score(y, y_pred))
```

### 4.2 决策树回归

```python
from sklearn.tree import DecisionTreeRegressor

# 创建模型
dt = DecisionTreeRegressor(max_depth=3)

# 训练
dt.fit(X_train, y_train)

# 预测
y_pred = dt.predict(X_test)
```

---

## 5. 无监督学习：聚类

### 5.1 K-Means聚类

```python
from sklearn.cluster import KMeans
import matplotlib.pyplot as plt

# 创建数据
X = np.array([[1, 1], [1, 2], [5, 5], [5, 6]])

# 创建K-Means模型
kmeans = KMeans(n_clusters=2, random_state=42)

# 训练
kmeans.fit(X)

# 预测聚类标签
labels = kmeans.labels_
centers = kmeans.cluster_centers_

print("聚类标签：", labels)
print("聚类中心：\n", centers)
```

### 5.2 层次聚类

```python
from sklearn.cluster import AgglomerativeClustering

# 创建模型
agg = AgglomerativeClustering(n_clusters=2)

# 训练
agg.fit(X)

# 预测
labels = agg.labels_
```

---

## 6. 模型评估与选择

### 6.1 交叉验证

```python
from sklearn.model_selection import cross_val_score

# 5折交叉验证
scores = cross_val_score(knn, X, y, cv=5)

print("每折得分：", scores)
print("平均得分：", scores.mean())
print("标准差：", scores.std())
```

### 6.2 网格搜索（超参数优化）

```python
from sklearn.model_selection import GridSearchCV

# 定义参数网格
param_grid = {
    'n_neighbors': [3, 5, 7, 9],
    'weights': ['uniform', 'distance']
}

# 创建网格搜索
grid_search = GridSearchCV(
    KNeighborsClassifier(),
    param_grid,
    cv=5,
    scoring='accuracy'
)

# 训练
grid_search.fit(X_train, y_train)

# 最佳参数
print("最佳参数：", grid_search.best_params_)
print("最佳得分：", grid_search.best_score_)

# 使用最佳模型预测
y_pred = grid_search.predict(X_test)
```

---

## 7. 实战项目：房价预测

### 7.1 数据加载

```python
import pandas as pd
from sklearn.datasets import fetch_california_housing

# 加载加州房价数据集
housing = fetch_california_housing()
X = pd.DataFrame(housing.data, columns=housing.feature_names)
y = housing.target

# 查看数据
print(X.head())
print(X.describe())
```

### 7.2 数据划分

```python
from sklearn.model_selection import train_test_split

# 划分数据集
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)
```

### 7.3 特征工程

```python
from sklearn.preprocessing import StandardScaler

# 标准化
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test)
```

### 7.4 模型训练

```python
from sklearn.ensemble import RandomForestRegressor
from sklearn.linear_model import LinearRegression
from sklearn.svm import SVR

# 多个模型
models = {
    'Linear Regression': LinearRegression(),
    'Random Forest': RandomForestRegressor(n_estimators=100, random_state=42),
    'SVR': SVR(kernel='rbf')
}

# 训练和评估
from sklearn.metrics import mean_squared_error, r2_score

for name, model in models.items():
    model.fit(X_train_scaled, y_train)
    y_pred = model.predict(X_test_scaled)
    
    mse = mean_squared_error(y_test, y_pred)
    r2 = r2_score(y_test, y_pred)
    
    print(f"{name}:")
    print(f"  MSE: {mse:.2f}")
    print(f"  R²: {r2:.2f}\n")
```

### 7.5 模型优化

```python
# 网格搜索优化随机森林
param_grid = {
    'n_estimators': [50, 100, 200],
    'max_depth': [None, 10, 20],
    'min_samples_split': [2, 5, 10]
}

grid_search = GridSearchCV(
    RandomForestRegressor(random_state=42),
    param_grid,
    cv=5,
    scoring='neg_mean_squared_error'
)

grid_search.fit(X_train_scaled, y_train)

print("最佳参数：", grid_search.best_params_)
print("最佳MSE：", -grid_search.best_score_)
```

---

## 8. 模型保存与加载

### 8.1 保存模型

```python
import joblib

# 保存模型
joblib.dump(grid_search.best_estimator_, 'house_price_model.pkl')
print("模型已保存！")
```

### 8.2 加载模型

```python
# 加载模型
loaded_model = joblib.load('house_price_model.pkl')

# 使用模型预测
new_data = [[8.3252, 41.0, 6.9841, 1.0238, 322.0, 2.55, 37.88, -122.23]]
prediction = loaded_model.predict(new_data)
print("预测房价：", prediction[0])
```

---

## 9. Pipeline完整流程

```python
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# 创建Pipeline
pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('classifier', RandomForestClassifier(n_estimators=100))
])

# 加载数据
iris = load_iris()
X, y = iris.data, iris.target

# 划分数据
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.3, random_state=42
)

# 训练
pipeline.fit(X_train, y_train)

# 预测
y_pred = pipeline.predict(X_test)

# 评估
print("准确率：", accuracy_score(y_test, y_pred))
```

---

## 10. 常见问题与解决

### 10.1 过拟合

**症状**：训练集准确率99%，测试集准确率70%

**解决方案**：
- 增加训练数据
- 使用正则化
- 减少模型复杂度
- 交叉验证

```python
from sklearn.linear_model import Ridge

# 使用正则化
ridge = Ridge(alpha=1.0)  # alpha是正则化强度
ridge.fit(X_train, y_train)
```

### 10.2 数据不平衡

**症状**：分类器总是预测多数类

**解决方案**：
- 调整类别权重
- 过采样/欠采样
- 使用合适的评估指标（F1-score）

```python
from sklearn.svm import SVC

# 设置类别权重
svc = SVC(class_weight='balanced')
svc.fit(X_train, y_train)
```

---

## 总结

**Scikit-learn核心要点**：
1. 数据预处理（标准化、编码）
2. 选择合适的模型
3. 训练与评估
4. 超参数优化
5. 模型保存与部署

**学习路径**：
1. 掌握基础API
2. 理解核心算法
3. 实战项目练习
4. 深入特定领域（NLP、CV）

**推荐资源**：
- 官方文档：https://scikit-learn.org/
- 《Python机器学习基础教程》
- Kaggle实战竞赛

**下一篇预告**：《VS Code插件推荐：提高效率50%》

---

**如果这篇文章对你有帮助，请点赞、收藏、转发！**  
**有问题欢迎评论区讨论！** 💪

---

持续更新机器学习实战教程，欢迎关注！