插入符号训练预处理中的相关截止

Question

我正在使用 r 中的 caret 包构建 C5.0 模型。

control <- trainControl(method = "repeatedcv", 
                    number = 10, 
                    repeats = 3, 
                    classProbs = TRUE, 
                    sampling = 'smote',
                    returnResamp="all",
                    summaryFunction = twoClassSummary)

grid <- expand.grid(.winnow = c(FALSE, TRUE), 
                 .trials = c(1, 5,10,15,20,25,30,40,45,50), 
                 .model= c("tree"),
                 .splits=c(2,5,10,15,20,25,50))

c5_model <- train(label ~ .,
              data = train,
              trControl = control, 
              method = c5info,
              tuneGrid = grid, 
              preProcess = c("center", "scale", "nzv","corr"),
              verbose = FALSE)

是否可以将自定义截止点传递给 preProcess 函数以进行相关性 - 例如 0.75 或我想要的任何点？

Answer 1

您可以在trainControl中指定预处理选项：

library(caret)
library(mlbench) #for the data
data(Sonar)

ctrl <-trainControl(method = "repeatedcv", 
                    number = 10, 
                    repeats = 3, 
                    classProbs = TRUE, 
                    sampling = 'smote',
                    returnResamp="all",
                    summaryFunction = twoClassSummary,
                    preProcOptions = list(cutoff = 0.75)) # all go in this list

一些游侠模型：

grid <- expand.grid(.mtry = c(2,5,10),
                    .min.node.size = 2,
                    .splitrule = "gini")

fit_model <- train(Class ~ .,
                  data = Sonar,
                  trControl = ctrl, 
                  metric = "ROC",
                  method = "ranger",
                  tuneGrid = grid,
                  preProcess = c("center", "scale", "nzv","corr"),
                  verbose = FALSE)

fit_model$preProcess
#output
Created from 679 samples and 60 variables

Pre-processing:
  - centered (26)
  - ignored (0)
  - removed (34)
  - scaled (26)

使用不同的截止值：

ctrl2 <-trainControl(method = "repeatedcv", 
                    number = 10, 
                    repeats = 3, 
                    classProbs = TRUE, 
                    sampling = 'smote',
                    returnResamp="all",
                    summaryFunction = twoClassSummary,
                    preProcOptions = list(cutoff = 0.6))

fit_model2 <- train(Class ~ .,
                   data = Sonar,
                   trControl = ctrl2, 
                   metric = "ROC",
                   method = "ranger",
                   tuneGrid = grid,
                   preProcess = c("center", "scale", "nzv","corr"),
                   verbose = FALSE)

fit_model2$preProcess
#output
Created from 679 samples and 60 variables

Pre-processing:
  - centered (23)
  - ignored (0)
  - removed (37)
  - scaled (23)

更多列被删除

当我们使用preProcOptions = list(cutoff = 0.95))

fit_model3$preProcess
#output
Created from 679 samples and 60 variables

Pre-processing:
  - centered (55)
  - ignored (0)
  - removed (5)
  - scaled (55)

看起来很有效。

同样，您可以传递任何其他预处理选项：

?caret::preProcess

检查所有的