from sklearn.ensemble import RandomForestRegressor #导入随机森林的包
import pandas as pd
#加载入数据，这里用的是住房的数据
from sklearn.datasets.california_housing import fetch_california_housing
housing = fetch_california_housing()
#进行数据的分割， test_size表示分割的比例
from sklearn.model_selection import train_test_split
data_train, data_test, target_train, target_test = \
train_test_split(housing.data, housing.target, test_size = 0.1, random_state = 42)
#构建随机森林的树
rfr = RandomForestRegressor(random_state=42) #每次生成的数都是相同的
rfr.fit(data_train, target_train) #模型计算
rfr_predict = rfr.predict(data_test) #预测结果
rfr.score(data_train, target_train) #默认评估值
#随机森林的参数调节
from sklearn.grid_search import GridSearchCV
tree_pram_grad = {'min_samples_split':list((3, 6, 9)), 'n_estimators':list((10,50,100))}
grid = GridSearchCV(RandomForestRegressor(), param_grid=tree_pram_grad, cv=5) #RandomForestRegressor()
# 表示需要调节的函数程序， param_grid 表示待调节的参数， cv=5表示交叉验证的次数
grid.fit(data_train, target_train)
print(grid.grid_scores_, grid.best_params_, grid.best_score_)
# 输出每组的grid_scores， 输出最好的参数组合，输出最好的默认评估值
# 输出的结果是在3 和 100 时呈现最好的状态
# 重新构建树做预测
rfr = RandomForestRegressor(random_state=42) #每次生成的数都是相同的
rfr.fit(data_train, target_train) #模型计算
new_rfr_predict = rfr.predict(data_test) #预测结果
#展示变量的重要性
pd.Series(rfr.feature_importances_, index=housing.feature_names).sort_values(ascending=False)
