不仅可视化决策树的训练集标签分布，还应该包括测试集标签分布。

Question

我们可以使用训练集分布来可视化决策树，例如。

from matplotlib import pyplot as plt
from sklearn import datasets
from sklearn.tree import DecisionTreeClassifier 
from sklearn import tree

# Prepare the data data, can do row sample and column sample here
iris = datasets.load_iris()
X = iris.data
y = iris.target
# Fit the classifier with default hyper-parameters
clf = DecisionTreeClassifier(random_state=1234)
clf.fit(X, y)


fig = plt.figure(figsize=(25,20))
_ = tree.plot_tree(clf, 
                   feature_names=iris.feature_names,  
                   class_names=iris.target_names,
                   filled=True)

让我们通过来了解训练集的分布情况，例如根节点中value = [50, 50, 50]。

然而，我无法为其提供测试集，并在可视化树中获得测试集的分布情况。

- Jiadong Chen

Answer 1

我认为目前还没有sklearn方法可以做到这一点。

选项1：通过添加X_test信息更改树的注释图

您可以使用以下自定义函数：

def plot_tree_test(clf, tree_plot, X_test, y_test):

    n = len(tree_plot)
    cat = clf.n_classes_

    # Getting the path for each item in X_test
    path = clf.decision_path(X_test).toarray().transpose()

    # Looping through each node/leaf in the tree and adding information from X_test path
    for i in range(n):
        value = []
        for j in range(cat):
            value += [sum(y_test[path[i]==1]==j)]
        tree_plot[i].set_text(tree_plot[i].get_text()+f'\ntest samples = {path[i].sum()}\ntest value = {value}')
    
    return tree_plot

稍微修改一下脚本：

from matplotlib import pyplot as plt
from sklearn import datasets
from sklearn.tree import DecisionTreeClassifier 
from sklearn import tree
from sklearn.model_selection import train_test_split

# Prepare the data data, can do row sample and column sample here
iris = datasets.load_iris()
X = iris.data
y = iris.target

# Creating a train and test set
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, shuffle=True, random_state=1234)

# Fit the classifier with default hyper-parameters
clf = DecisionTreeClassifier(random_state=1234)
clf.fit(X_train, y_train)

fig = plt.figure(figsize=(25,20))
tree_plot = tree.plot_tree(clf, 
                   feature_names=iris.feature_names,  
                   class_names=iris.target_names,
                   filled=True)

tree_plot = plot_tree_test(clf, tree_plot, X_test, y_test)
plt.show()

输出：

选项2：使用X_test信息更改分类器本身

您可以使用以下自定义函数：

def tree_test(clf, X_test, y_test):

    state = clf.tree_.__getstate__()
    n = len(state['values'])
    cat = clf.n_classes_
    
    # Getting the path for each item in X_test
    path = clf.decision_path(X_test).toarray().transpose()
    
    # Looping through each node/leaf in the tree and adding information from X_test path
    values = []
    for i in range(n):
        value = []
        for j in range(cat):
            value += [float(sum(y_test[path[i]==1]==j))]
        values += [[value]]
        state['nodes'][i][5] = path[i].sum()
        state['nodes'][i][6] = max(path[i].sum(), 0.1) # 0 returns error
    
    values = np.array(values)
    state['values'] = values
    clf.tree_.__setstate__(state)
    return clf

稍微更改脚本：

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

# Prepare the data data, can do row sample and column sample here
iris = datasets.load_iris()
X = iris.data
y = iris.target

# Creating a train and test set
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, shuffle=True, random_state=1234)

# Fit the classifier with default hyper-parameters
clf = DecisionTreeClassifier(random_state=1234)
clf.fit(X_train, y_train)

clf = tree_test(clf, X_test, y_test)

fig = plt.figure(figsize=(25,20))
tree_plot = tree.plot_tree(clf, 
                   feature_names=iris.feature_names,  
                   class_names=iris.target_names,
                   filled=True)

plt.show()

输出：

- Mattravel