决策树之剪枝介绍

import numpy as np
import matplotlib.pyplot as plt
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.datasets import make_classification

# 1. 生成数据
X, y = make_classification(
    n_samples=1000,
    n_features=20,
    n_informative=5,
    n_redundant=5,
    n_classes=2,
    random_state=42
)

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.3, random_state=42
)

print(f"训练集: {X_train.shape}")
print(f"测试集: {X_test.shape}")

# 2. 不剪枝的决策树（过拟合）
print("\n=== 不剪枝的决策树 ===")
dt_no_prune = DecisionTreeClassifier(random_state=42)
dt_no_prune.fit(X_train, y_train)

train_acc = accuracy_score(y_train, dt_no_prune.predict(X_train))
test_acc = accuracy_score(y_test, dt_no_prune.predict(X_test))
print(f"训练集准确率: {train_acc:.4f}")
print(f"测试集准确率: {test_acc:.4f}")
print(f"树的深度: {dt_no_prune.get_depth()}")
print(f"叶子节点数: {dt_no_prune.get_n_leaves()}")

# 3. 预剪枝（限制参数）
print("\n=== 预剪枝（限制参数）===")
dt_pre_prune = DecisionTreeClassifier(
    max_depth=5,              # 最大深度
    min_samples_split=10,     # 内部节点再划分最小样本数
    min_samples_leaf=5,        # 叶子节点最少样本数
    random_state=42
)
dt_pre_prune.fit(X_train, y_train)

train_acc_pre = accuracy_score(y_train, dt_pre_prune.predict(X_train))
test_acc_pre = accuracy_score(y_test, dt_pre_prune.predict(X_test))
print(f"训练集准确率: {train_acc_pre:.4f}")
print(f"测试集准确率: {test_acc_pre:.4f}")
print(f"树的深度: {dt_pre_prune.get_depth()}")
print(f"叶子节点数: {dt_pre_prune.get_n_leaves()}")

# 4. 后剪枝（ccp_alpha参数）
print("\n=== 后剪枝（ccp_alpha）===")
dt_post_prune = DecisionTreeClassifier(
    ccp_alpha=0.01,  # 复杂度参数，越大剪枝越多
    random_state=42
)
dt_post_prune.fit(X_train, y_train)

train_acc_post = accuracy_score(y_train, dt_post_prune.predict(X_train))
test_acc_post = accuracy_score(y_test, dt_post_prune.predict(X_test))
print(f"训练集准确率: {train_acc_post:.4f}")
print(f"测试集准确率: {test_acc_post:.4f}")
print(f"树的深度: {dt_post_prune.get_depth()}")
print(f"叶子节点数: {dt_post_prune.get_n_leaves()}")

# 5. 对比不同ccp_alpha
print("\n=== 不同ccp_alpha对比 ===")
ccp_alphas = [0, 0.001, 0.01, 0.05, 0.1]
train_accs = []
test_accs = []
depths = []
n_leaves = []

for alpha in ccp_alphas:
    dt = DecisionTreeClassifier(ccp_alpha=alpha, random_state=42)
    dt.fit(X_train, y_train)
    
    train_accs.append(accuracy_score(y_train, dt.predict(X_train)))
    test_accs.append(accuracy_score(y_test, dt.predict(X_test)))
    depths.append(dt.get_depth())
    n_leaves.append(dt.get_n_leaves())
    
    print(f"alpha={alpha:.3f}: 训练={train_accs[-1]:.4f}, 测试={test_accs[-1]:.4f}, "
          f"深度={depths[-1]}, 叶子={n_leaves[-1]}")

# 6. 可视化
plt.figure(figsize=(15, 5))

# 准确率对比
plt.subplot(1, 3, 1)
plt.plot(ccp_alphas, train_accs, 'bo-', label='训练集')
plt.plot(ccp_alphas, test_accs, 'ro-', label='测试集')
plt.xlabel('ccp_alpha')
plt.ylabel('准确率')
plt.title('准确率 vs ccp_alpha')
plt.legend()
plt.grid(True)

# 深度对比
plt.subplot(1, 3, 2)
plt.plot(ccp_alphas, depths, 'go-')
plt.xlabel('ccp_alpha')
plt.ylabel('树的深度')
plt.title('树的深度 vs ccp_alpha')
plt.grid(True)

# 叶子节点数对比
plt.subplot(1, 3, 3)
plt.plot(ccp_alphas, n_leaves, 'mo-')
plt.xlabel('ccp_alpha')
plt.ylabel('叶子节点数')
plt.title('叶子节点数 vs ccp_alpha')
plt.grid(True)

plt.tight_layout()
plt.show()

# 7. 三种情况对比
print("\n=== 三种情况对比 ===")
print(f"{'方法':<15} {'训练准确率':<12} {'测试准确率':<12} {'深度':<8} {'叶子数':<8}")
print("-" * 55)
print(f"{'不剪枝':<15} {train_acc:<12.4f} {test_acc:<12.4f} {dt_no_prune.get_depth():<8} {dt_no_prune.get_n_leaves():<8}")
print(f"{'预剪枝':<15} {train_acc_pre:<12.4f} {test_acc_pre:<12.4f} {dt_pre_prune.get_depth():<8} {dt_pre_prune.get_n_leaves():<8}")
print(f"{'后剪枝':<15} {train_acc_post:<12.4f} {test_acc_post:<12.4f} {dt_post_prune.get_depth():<8} {dt_post_prune.get_n_leaves():<8}")