Model request: bam models from mgcv

[spedygiorgio]

Mgcv package has recently proposed an optimized version of gam... It could be nice to add this feature https://stat.ethz.ch/R-manual/R-devel/library/mgcv/html/bam.html

[topepo]

Go ahead and try that code out for testing.

[topepo]

There seems to be an error:
Error in train.default(x = dat[, c("x1", "x2", "x3", "x0")], y = dat$y, :
Model bam is not in caret's built-in library

Also I would suggest to let the family not be bounded to binomial or gaussian cause the model could be very useful to fit other glms (poisson or gamma ones for example)

It isn't in the package yet. You'll have to source the model code into an R session then use method = modelInfo. For example:

library(caret)
set.seed(24)
tr_dat <- SLC14_1(500)

set.seed(13)
mod <- train(y ~ ., data = tr_dat, 
             method = modelInfo,
             trControl = trainControl(method = "cv"))

You should also be able to pass family in also. For example:

hist(exp(tr_dat$y/100))

set.seed(13)
mod <- train(exp(y/100) ~ ., data = tr_dat, 
             method = modelInfo,
             family = Gamma(),
             trControl = trainControl(method = "cv"))

[topepo]

on CRAN