problem with local variables and gbm formula interface

[patr1ckm]

I ran into the following problem where gbm does not find a predictor in the local environment:

do.gbm <- function(y,x){
  require(gbm)
  l <- x
  o <- gbm(y~l)
}

x <- rnorm(100)
y <- x^2
do.gbm(y,x)

Produces

 Error in eval(expr, envir, enclos) : object 'l' not found 

It occurs at the following line (456), which looks for the variables within data rather than parent.frame():

   x <- model.frame(terms(reformulate(var.names)),
                    data,
                    na.action=na.pass,
                    subset=subset)

A proposed solution could be to just select the predictors from mf using var.names computed the line before. mf should have all variables either in data or parent.frame. I've not tested this extensively.

x <- mf[,var.names,drop=F]

[pdmetcalfe]

Yes. It also has problems with gbm((fred > 0) ~ ., data=stuff). Fixing this is (in my head) kind of tied in with refactoring the R code to remove all the if else from gbm, gbm.fit, etc.