Question

我试图比较几种回归和分类算法的性能，但是“e1071”包中的svm函数不会按照我喜欢的方式工作。

所以这是我的代码：

library("mlbench")   
library("e1071")
library("hydroGOF")

data(BostonHousing)
data = BostonHousing
rm(BostonHousing)
data = na.omit(data)

svm = function(j, data){
 ### some preparations
 n = dim(data)[1]
 ### empty vector for CV                                          
 fold_j = numeric(n)
 ### assign numbers for every observation from 1 to 10 for CV                                       
 fold_svm = c(rep(1:10, floor(n/10)), 1:(n%%10))
 ### empty vector for infold CV             
 MSE_infold = numeric(10)  
 ### empty vector for cross validated MSE                     
 MSE = numeric(j)  
  for(j in 1:j){
   print(paste0("SVM ", j))
   ### pick a sample for 10-fold CV
   fold_j = sample(fold_svm, n, replace = FALSE)
    for(i in 1:10){
     print(paste0("CV ", i))
     ### define train data 
     traindat_svm = data[fold_j != i, ]
     ### define test data
     newdat = data[fold_j == i, ]
     ### svm
     svm_model = svm(medv ~ ., 
                     data = traindat_svm, 
                     cost = 1000, 
                     gamma = 0.0001)
     ### predict values for response (medv = column 14) with svm
     svm_pred = predict(svm_model, newdat[,-14])
     ### calculate standardized MSE infold
     MSE_infold[i] = ((mse(svm_pred, newdat[,14])) / var(data$medv)) * 100
    }
  ### calculate MSE
  MSE[j] = mean(MSE_infold)
  }
 MSE
}

如果我使用我的功能会发生什么，比如说

svm(10, data)

是我收到以下错误消息：

Error in svm(medv ~ ., data = traindat_svm, cost = 1000, gamma = 1e-04) : 
  unused arguments (cost = 1000, gamma = 1e-04)

如果我单独执行每个命令而没有函数/ for循环，它可以很好地工作。

任何想法？

提前致谢，Niklas

交叉验证的SVM作为函数不运行

0 个答案: