Currently, the `gam' method gives an error if passed with trainControl(method = "none"). You can see the error here:
rm( list = ls())
n <- 1e3
x <- cbind(x = rnorm(n))
y <- c(x + rnorm(n))
train(x, y, method = "gam", trControl = trainControl(method = "none"))
# Error: Only one model should be specified in tuneGrid with no resampling
Note that the default behavior of trainControl(method = "none") is to set tuneLength = 1 and thus create a grid of length 1.
However, here the minimal grid it was creating always had length 2, as it was passing both TRUE and FALSE to the parameter `select,' which does not make sense if we want a grid of length 1.
The proposed change makes sure expand.grid will always have length 1, if len = 1.
(and this does not affect the behavior of anything else whenever len > 1)
topepo
commented
1 year ago
Currently, the `gam' method gives an error if passed with trainControl(method = "none"). You can see the error here:
Note that the default behavior of
trainControl(method = "none")is to settuneLength = 1and thus create a grid of length 1.However, here the minimal grid it was creating always had length 2, as it was passing both TRUE and FALSE to the parameter `select,' which does not make sense if we want a grid of length 1.
The proposed change makes sure expand.grid will always have length 1, if len = 1.
(and this does not affect the behavior of anything else whenever len > 1)