我无法运行插入符号的 nnet 回归

Question

我尝试使用 caret 的 nnet 运行回归，但出现错误。

library(tidyverse)
library(caret)
feature = rnorm(100, 0, 1) %>% as.matrix() 
colnames(feature) = "x1"
outcome = rnorm(100, 0, 1) %>% as.matrix()
colnames(feature) = "y1"
model = caret::train(
  x = feature, y = outcome, method = "nnet",
  tuneGrid = expand.grid(size=c(1:3), decay=seq(0.1, 1, 0.1)),
  weights = NULL, linout = TRUE
  )

Error: Metric RMSE not applicable for classification models

当然，我想回归，而不是分类。为了显示这一点，我设置了选项linout = TRUE。出了什么问题？

另外，我关注this question，并试图删除as.matrix，但它也显示其他错误。

library(tidyverse)
library(caret)
feature = rnorm(100, 0, 1) %>% as.double()
outcome = rnorm(100, 0, 1) %>% as.double()
CATE_model = caret::train(
  x = feature, y = outcome, method = "nnet",
  tuneGrid = expand.grid(size=c(1:3), decay=seq(0.1, 1, 0.1)),
  weights = NULL, linout = TRUE
  )

Error: Please use column names for 'x'

非常感谢

Answer 1

神经网络的输入似乎需要列名。 这是有道理的，因为经过训练的神经网络稍后可能需要识别其输入中的正确列以进行预测。

如果您将您的功能强制放入带有 colnames 的 Dataframe（在我的情况下为 x），它会起作用。

library(tidyverse)
library(caret)
feature = data.frame(x = rnorm(100, 0, 1) %>% as.double()) # changed line 
outcome = rnorm(100, 0, 1) %>% as.double()
CATE_model = caret::train(
  x = feature, y = outcome, method = "nnet",
  tuneGrid = expand.grid(size=c(1:3), decay=seq(0.1, 1, 0.1)),
  weights = NULL, linout = TRUE
)