警告消息：在 method$prob(modelFit =modelFit, newdata =newdata, submodels =param) 中： kernlab 类概率计算失败；返回 NA

Question

我收到上面的以下错误，该错误似乎与插入符号包有关。我检查了我的数据，没有任何 NA 值，因为它们在清理阶段全部被删除。 我正在尝试获取 AUC 的概率，以便与我在项目中使用的其他 ML 算法进行比较。我的代码如下：

    Cost = 2^c(1:8)

    svm.control = trainControl(
    method = "cv", # use cross validation
    number = 10, # with 10 folds,
    summaryFunction = defaultSummary,
    #classProbs = TRUE
    )


   svm.linear.grid <- expand.grid(
         C = Cost
   )


   svm.fit1 <- train( # Train a model model
       final_result ~ ., 
       data = sd.BBB.train.rose, # sd.BBB.train.rose as the training data
       method = "svmLinear", # using the linear kernel
       trControl = svm.control, # cross-validated as configured above
       preProc = c("center", "scale","nzv"), 
       verbose = FALSE,
       #metric="ROC",
       tuneGrid = svm.linear.grid # use the tuning grid created above
   )

   #roc and auc analysis 
   svm.prob <- predict(svm.fit2, newdata = sd.BBB.test[,-8], type="prob")

我的数据集的结构

  > str(sd.BBB.train.rose)
  'data.frame': 4926 obs. of  8 variables:
  $ gender              : num  0.0713 -0.0131 0.0321 -0.0881 -0.0344 ...
  $ highest_education   : num  0.0761 0.6618 0.2545 1.275 -0.2358 ...
  $ age_group           : num  -0.1609 0.023 0.0563 0.103 -0.2153 ...
  $ previous_attempts   : num  -0.0689 0.8812 -0.2724 0.9355 0.9366 ...
  $ disability          : num  -0.0664 0.2857 -0.0461 1.0651 -0.0573 ...
  $ avg_clicks          : num  0.588 4.666 4.174 3.228 3.532 ...
  $ academic_performance: num  3.95 1.39 3.51 -1.84 4.31 ...
  $ final_result        : Factor w/ 2 levels "0","1": 1 1 1 1 1 1 1 1 1 1 ...
 > 

非常感谢任何帮助。尊敬。

Answer 1

如果没有可用的数据来重现错误，很难确定，但我认为如果您想获得返回的概率，则必须在 train 中指定 classProbs = TRUE代码>函数。试试这个：

svm.fit1 <- train( # Train a model model
       final_result ~ ., 
       data = sd.BBB.train.rose, # sd.BBB.train.rose as the training data
       method = "svmLinear", # using the linear kernel
       trControl = svm.control, # cross-validated as configured above
       preProc = c("center", "scale","nzv"), 
       verbose = FALSE,
       #metric="ROC",
       tuneGrid = svm.linear.grid, # use the tuning grid created above
       classProbs = TRUE
   )