r - 如何将 ntree 参数强加到 caret 包的 train() 函数中？

Question

我正在使用以下函数在我的数据集上与随机森林算法进行交叉验证。但是，ntree 会引发错误，指出函数中未使用它。尽管我之前在有关此问题的一个线程中看到过这种用法作为推荐评论，但它对我不起作用。这是我的代码:

cv_rf_class1 <- train(y_train_u ~ ., x_train_u , 
                      method ="cforest", 
                      trControl = trainControl(method = "cv", 
                                               number = 10, 
                                               verboseIter = TRUE),  
                                               ntree = 100))

如果我无法更改 ntree 参数，它会在函数中使用 500 棵树作为默认值，并且会给我带来另一个错误(下标越界)，因此我无法让它解决我的问题。如何解决此问题以使我的功能正常工作？

Answer 1

ntree 需要是 train 的参数，而不是您在此处使用的 trainControl 的参数；来自 train 的文档:

...
arguments passed to the classification or regression routine (such as randomForest). Errors will occur if values for tuning parameters are passed here.

另请注意，您没有以正确的形式传递数据； train 期望数据为 (x, y)，而不是您传递的数据(公式和矩阵的不正确组合)。

总而言之，将您的 train 调用更改为:

cv_rf_class1 <- train(x_train_u, y_train_u,
                      method ="cforest", 
                      ntree = 100,
                      trControl = trainControl(method = "cv", 
                                               number = 10, 
                                               verboseIter = TRUE))

更新(评论后)

好吧，似乎 cforest 特别不会接受 ntree 参数，因为与原始的 randomForest 包相比，这个不是您在相应包 ( docs ) 的底层 cforest 函数中传递树数量的方式。

正确的方法，如 caret Github repo 中的相关示例所示，是:

cv_rf_class1 <- train(x_train_u, y_train_u,
                      method ="cforest", 
                      trControl = trainControl(method = "cv", 
                                               number = 10, 
                                               verboseIter = TRUE),
                      controls = party::cforest_unbiased(ntree = 100))

适应cforest.R ，我们得到:

library(caret)
library(plyr)
library(recipes)
library(dplyr)

model <- "cforest"

set.seed(2)
training <- twoClassSim(50, linearVars = 2)
testing <- twoClassSim(500, linearVars = 2)
trainX <- training[, -ncol(training)]
trainY <- training$Class

rec_cls <- recipe(Class ~ ., data = training) %>%
  step_center(all_predictors()) %>%
  step_scale(all_predictors())

seeds <- vector(mode = "list", length = nrow(training) + 1)
seeds <- lapply(seeds, function(x) 1:20)

cctrl1 <- trainControl(method = "cv", number = 3, returnResamp = "all",
                       classProbs = TRUE, 
                       summaryFunction = twoClassSummary,
                       seeds = seeds)

set.seed(849)
test_class_cv_model <- train(trainX, trainY, 
                             method = "cforest", 
                             trControl = cctrl1,
                             metric = "ROC", 
                             preProc = c("center", "scale"),
                             controls = party::cforest_unbiased(ntree = 20)) # WORKS OK

test_class_pred <- predict(test_class_cv_model, testing[, -ncol(testing)])
test_class_prob <- predict(test_class_cv_model, testing[, -ncol(testing)], type = "prob")

head(test_class_pred)
# [1] Class2 Class2 Class2 Class1 Class1 Class1
# Levels: Class1 Class2

head(test_class_prob)
#      Class1    Class2
# 1 0.4996686 0.5003314
# 2 0.4333222 0.5666778
# 3 0.3625118 0.6374882
# 4 0.5373396 0.4626604
# 5 0.6174159 0.3825841
# 6 0.5327283 0.4672717

sessionInfo()的输出:

R version 3.6.1 (2019-07-05)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1

Matrix products: default

locale:
[1] LC_COLLATE=English_United Kingdom.1252  LC_CTYPE=English_United Kingdom.1252    LC_MONETARY=English_United Kingdom.1252
[4] LC_NUMERIC=C                            LC_TIME=English_United Kingdom.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] recipes_0.1.7   dplyr_0.8.3     plyr_1.8.4      caret_6.0-84    ggplot2_3.2.1   lattice_0.20-38