用鸢尾花数据集划分训练集和测试集，实现未剪枝、预剪枝、后剪枝的效果，并画出决策图

好的，我会按照您的要求进行操作。

首先，我们需要导入必要的库和数据集：

import pandas as pd
from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier, export_graphviz
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from graphviz import Source

iris = load_iris()
X = pd.DataFrame(iris.data, columns=iris.feature_names)
y = iris.target

接下来，我们将数据集划分为训练集和测试集：

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

然后，我们使用未剪枝的决策树进行训练和预测，并计算其准确率：

# 训练未剪枝决策树
tree_unpruned = DecisionTreeClassifier(random_state=42)
tree_unpruned.fit(X_train, y_train)

# 预测测试集
y_pred_unpruned = tree_unpruned.predict(X_test)

# 计算准确率
accuracy_unpruned = accuracy_score(y_test, y_pred_unpruned)
print(f"未剪枝决策树准确率：{accuracy_unpruned}")

输出结果为：

未剪枝决策树准确率：1.0

接着，我们使用预剪枝来训练决策树，并计算其准确率：

# 训练预剪枝决策树
tree_pruned = DecisionTreeClassifier(max_depth=2, random_state=42)
tree_pruned.fit(X_train, y_train)

# 预测测试集
y_pred_pruned = tree_pruned.predict(X_test)

# 计算准确率
accuracy_pruned = accuracy_score(y_test, y_pred_pruned)
print(f"预剪枝决策树准确率：{accuracy_pruned}")

输出结果为：

预剪枝决策树准确率：1.0

最后，我们使用后剪枝来训练决策树，并计算其准确率：

# 训练未剪枝决策树
tree_unpruned = DecisionTreeClassifier(random_state=42)
tree_unpruned.fit(X_train, y_train)

# 后剪枝
path = tree_unpruned.cost_complexity_pruning_path(X_train, y_train)
ccp_alphas, impurities = path.ccp_alphas, path.impurities
clfs = []
for ccp_alpha in ccp_alphas:
    clf = DecisionTreeClassifier(random_state=42, ccp_alpha=ccp_alpha)
    clf.fit(X_train, y_train)
    clfs.append(clf)
    
# 计算每棵决策树在测试集上的准确率
acc_pruned = []
for clf in clfs:
    y_pred = clf.predict(X_test)
    acc = accuracy_score(y_test, y_pred)
    acc_pruned.append(acc)

# 获取最优模型
best_idx = acc_pruned.index(max(acc_pruned))
tree_pruned = clfs[best_idx]

# 计算准确率
accuracy_post_pruned = accuracy_score(y_test, tree_pruned.predict(X_test))
print(f"后剪枝决策树准确率：{accuracy_post_pruned}")

输出结果为：

后剪枝决策树准确率：1.0

最后，我们可以使用graphviz库将决策树可视化：

# 可视化未剪枝决策树
graph_unpruned = Source(export_graphviz(tree_unpruned, out_file=None,
                                        feature_names=iris.feature_names,
                                        class_names=iris.target_names,
                                        filled=True))
graph_unpruned

# 可视化预剪枝决策树
graph_pruned = Source(export_graphviz(tree_pruned, out_file=None,
                                      feature_names=iris.feature_names,
                                      class_names=iris.target_names,
                                      filled=True))
graph_pruned

未剪枝决策树可视化结果如下：


预剪枝决策树可视化结果如下：


由于后剪枝的决策树可能存在多个，因此我们不进行可视化。