转：lightGBM的黑科技--plot函数

# -*- coding: utf-8 -*-
# @Time    : 2018/6/11
# @Author  : Reynold chen

import lightgbm as lgb
import numpy as np
import matplotlib.pyplot as plt

print(‘制造数据...‘)
x_train = np.random.random((1000, 10))
y_train = np.random.rand(1000) > 0.5
x_test = np.random.random((100, 10))
y_test = np.random.randn(100) > 0.5

# 导入到lightgbm矩阵
lgb_train = lgb.Dataset(x_train, y_train)
lgb_test = lgb.Dataset(x_test, y_test, reference=lgb_train)

# 设置参数
params = {
    ‘num_leaves‘: 5,
    ‘metric‘: (‘auc‘, ‘logloss‘),  # 可以设置多个评价指标
    ‘verbose‘: 0
}
# if (evals_result and gbm) not in locbals():
# global evals_result,gbm
# 如果是局部变量的话，推荐把他们变成全局变量，这样plot的代码位置不受限制
evals_result = {}  # 记录训练结果所用

print(‘开始训练...‘)
# train
gbm = lgb.train(params,
                lgb_train,
                num_boost_round=100,
                valid_sets=[lgb_train, lgb_test],
                evals_result=evals_result,  # 非常重要的参数,一定要明确设置,输出的结果是上面一个参数valid_sets配置的值
                verbose_eval=10)

print(evals_result)
print(‘画出训练结果...‘)
ax = lgb.plot_metric(evals_result, metric=‘auc‘)  # metric的值与之前的params里面的值对应
plt.show()

print(‘画特征重要性排序...‘)
lgb.plot_importance(gbm, max_num_features=10)  # max_features表示最多展示出前10个重要性特征，可以自行设置
plt.show()

print(‘Plot 3th tree...‘)  # 画出决策树，其中的第三颗
lgb.plot_tree(gbm, tree_index=3, figsize=(20, 8), show_info=[‘split_gain‘])
plt.show()

print(‘导出决策树的pdf图像到本地‘)  # 这里需要安装graphviz应用程序和python安装包
graph = lgb.create_tree_digraph(gbm, tree_index=3, name=‘Tree3‘)
graph.render(view=True)