EL之GB(GBC)：利用GB对多分类问题进行建模(分层抽样+调1参)并评估

EL之GB(GBC)：利用GB对多分类问题进行建模(分层抽样+调1参)并评估

目录

输出结果

设计思路

核心代码

输出结果

T1、

T2、

设计思路

核心代码

  

   

    

     

    
    

     

      #T1、
     
    
   

    

     

    
    

     

      nEst = 500
     
    
   

    

     

    
    

     

      depth = 3
     
    
   

    

     

    
    

     

      learnRate = 0.003
     
    
   

    

     

    
    

     

      maxFeatures = None
     
    
   

    

     

    
    

     

      subSamp = 0.5
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     

      #T2、
     
    
   

    

     

    
    

     

      # nEst = 500
     
    
   

    

     

    
    

     

      # depth = 3
     
    
   

    

     

    
    

     

      # learnRate = 0.003
     
    
   

    

     

    
    

     

      # maxFeatures = 3
     
    
   

    

     

    
    

     

      # subSamp = 0.5
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     

      glassGBMModel = ensemble.GradientBoostingClassifier(n_estimators=nEst, max_depth=depth,
     
    
   

    

     

    
    

     

       learning_rate=learnRate, max_features=maxFeatures,
     
    
   

    

     

    
    

     

       subsample=subSamp)
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     

      glassGBMModel.fit(xTrain, yTrain)
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     

      missClassError = []
     
    
   

    

     

    
    

     

      missClassBest = 1.0
     
    
   

    

     

    
    

     

      predictions = glassGBMModel.staged_decision_function(xTest)
     
    
   

    

     

    
    

     

      for p in predictions:
     
    
   

    

     

    
    

     

       missClass = 0
     
    
   

    

     

    
    

     
 for i in range(len(p)):
     
    
   

    

     

    
    

     

       listP = p[i].tolist()
     
    
   

    

     

    
    

     
 if listP.index(max(listP)) != yTest[i]:
     
    
   

    

     

    
    

     

       missClass += 1
     
    
   

    

     

    
    

     

       missClass = float(missClass)/len(p)
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     

       missClassError.append(missClass)
     
    
   

    

     

    
    

     
 
     
    
   

    

     

    
    

     
 if missClass < missClassBest:
     
    
   

    

     

    
    

     

       missClassBest = missClass
     
    
   

    

     

    
    

     

       pBest = p
     
    
  
 

文章来源: yunyaniu.blog.csdn.net，作者：一个处女座的程序猿，版权归原作者所有，如需转载，请联系作者。

原文链接：yunyaniu.blog.csdn.net/article/details/86442102