树模型分裂节点可视化

环境

• 不管用哪种可视化的模型，都需要配置graphviz图形驱动用来对结果进行展示，其他的安装包如下所示：

import pydotplus
from sklearn import tree
from IPython.display import Image

二叉树可视化代码

• 具体的操作代码如下所示：

import os

from sklearn import tree
from sklearn.datasets import load_iris

iris = load_iris()
clf = tree.DecisionTreeClassifier()
clf = clf.fit(train_x, train_y)

import pydotplus
from IPython.display import Image
dot_data = tree.export_graphviz(clf, out_file=None, 
                         feature_names=df_all.columns.tolist()[:-1],  
                         class_names=['0','1','2'], 
                         filled=True, rounded=True, special_characters=False)  
graph = pydotplus.graph_from_dot_data(dot_data)
graph.set('label', 'Tree 30 from  Xgboost Tree')
graph.set('labelloc', 't')
# 遍历节点
# 获取图中所有节点
nodes = graph.get_node_list()
for index, node in enumerate(nodes):
    # 判断节点名称是否为'node'，即右上方背景框
    if node.get_name().startswith('node'):
        # 设置节点样式和属性
        
        node.set('style', 'filled')
        # node.set('color', 'none')
Image(graph.create_png())
graph.write_png('decision_tree.png')

xgboost可视化代码

def model_plot_trees(clf, features, **n):
    """绘制树模型"""
    # 保存特征名称到fmap文件，用于图形绘制
    with open('xgb.fmap', 'w', encoding="utf-8") as fmap:
        for k, ft in enumerate(features):
            fmap.write(''.join(str([k, ft, 'p']) + '\n'))

    # 构建条件节点和叶子节点的形状、填充颜色
    c_node_params = {'shape': 'box',
                     'style': 'filled,rounded',
                     'fillcolor': '#78bceb'
                     }
    l_node_params = {'shape': 'box',
                     'style': 'filled',
                     'fillcolor': '#e48038'
                     }

    # 树模型绘制：有向图
    # 绘制和保存第num_trees+1棵树，num_trees为树的序号
    digraph = xgb.to_graphviz(clf, num_trees=30, condition_node_params=c_node_params,
                              leaf_node_params=l_node_params, fmap='xgb.fmap')
    # digraph.format = 'png'
    digraph.view('./oil_xgb_trees')

    # 分别绘制子图，不保存
    for i in range(n.get('n')):
        xgb.plot_tree(clf, num_trees=i, condition_node_params=c_node_params,
                      leaf_node_params=l_node_params, fmap='xgb.fmap')
        plt.show()
model_plot_trees(xgb_model, df_all.columns.tolist()[:-1])