python+GBDT实现数值预测

本是古典何须时尚 2022-04-16 02:27 266阅读 0赞

转载：[https://blog.csdn.net/tuntunwang/article/details/66969726][https_blog.csdn.net_tuntunwang_article_details_66969726]

这里展示如何对GBDT进行模型的训练，对数据进行预测(并非是分类)，对返回值进行处理，评价模型好坏

背景是天池的IJICAI大赛，数据已经是处理好的，这里直接用

代码如下

\#$encoding=utf-8  
'''  
环境 ubuntu+IDEA+python35  
实现的功能：利用GBDT模型实现数值的预测  
背景：天池的IJICAI，预测商店流量  
PS：feature\_data.csv是已经处理好的特征  
'''

代码如下：  
import numpy as np  
import pandas as pd  
from sklearn import ensemble, cross\_validation  
   
\#该评价指标用来评价模型好坏  
def rmspe(zip\_list,count):  
    \# w = ToWeight(y)  
    \# rmspe = np.sqrt(np.mean((y - yhat) \*\* 2))  
    sum\_value=0.0  
    \# count=len(zip\_list)  
    for real,predict in zip\_list:  
        v1=(real-predict)\*\*2  
        sum\_value += v1  
    v2=sum\_value / count  
    v3=np.sqrt(v2)  
    return v3  
   
\#提取特征和目标值  
def get\_features\_target(data):  
    data\_array=pd.np.array(data)\#传入dataframe，为了遍历，先转为array  
    features\_list=\[\]  
    target\_list=\[\]  
    for line in data\_array:  
        temp\_list=\[\]  
        for i in range(0,384):\#一共有384个特征  
            if i == 360 :\#index=360对应的特征是flow  
                target\_temp=int(line\[i\])  
            else:  
                temp\_list.append(int(line\[i\]))  
        features\_list.append(temp\_list)  
        target\_list.append(target\_temp)  
    return features\_list, target\_list  
    \# return pd.DataFrame(features\_list),pd.DataFrame(target\_list)  
   
def run\_demo():  
   
    feature\_save\_path = "/home/wangtuntun/IJCAI/Data/feature\_data.csv"  \# 将最终生成的特征存入该文件  
    data = pd.read\_csv(feature\_save\_path)  
    data\_other,data=cross\_validation.train\_test\_split(data,test\_size=0.001,random\_state=10)\#为了减少代码运行时间，方便测试  
    train\_and\_valid, test = cross\_validation.train\_test\_split(data, test\_size=0.2, random\_state=10)  
    train, valid = cross\_validation.train\_test\_split(train\_and\_valid, test\_size=0.01, random\_state=10)  
    train\_feature, train\_target = get\_features\_target(train)  
    test\_feature, test\_target = get\_features\_target(test)  
    valid\_feature, valid\_target = get\_features\_target(valid)  
   
    params = \{'n\_estimators': 500, 'max\_depth': 4, 'min\_samples\_split': 2,  
              'learning\_rate': 0.01, 'loss': 'ls'\}  
    clf = ensemble.GradientBoostingRegressor(\*\*params)  
   
    clf.fit(train\_feature, train\_target) \#训练  
    \# mse = mean\_squared\_error(test\_target, clf.predict(test\_feature)) \#预测并且计算MSE  
    \# print(mse)  
    pre=clf.predict(test\_feature)  
    pre\_list=list(pre)  
    real\_pre\_zip=zip(test\_target,pre\_list)  
   
    count=len(pre\_list)  
    error=rmspe(real\_pre\_zip,count)  
    print(error)  
   
run\_demo()

[https_blog.csdn.net_tuntunwang_article_details_66969726]: https://blog.csdn.net/tuntunwang/article/details/66969726