Skip to content

frames and formulas

2 messages · Kevin Murphy, Peter Dalgaard

Original
Kevin Murphy
# 
I can build a tree at the command line using
  tr <- tree(V4 ~ ., dat)
but I don't know how to build one inside a function - the following does
not work

  foo <- function(dummy) {
    tr <- tree(V4 ~ ., dummy)
  }
  tr2 <- foo(dat)

and produces the error

  Error in model.frame.default(formula = V4 ~ ., data = dummy) : 
	Object "dummy" not found

I know the solution has something to do with terms, formulas, frames,
environments, etc., but I can't figure it out (despite having read sec
3.4-3.5 of 'S progamming').

The reason I want to call 'tree' inside a function is that I am defining
a tree-based Conditional Probability Distribution (CPD) class that
should be able to be fit given an arbitrary number of parents. Currently
I implement this as shown below. This of course fails for the reason
above, but I also wondered if there was a way to avoid enumerating every
possible number of parents (I'm not sure how to make the formula "last
column ~ all other columns" in a general way).

fit.CPD.tree <- function(CPD, dat) {
  print(paste("fitting tree CPD", CPD$id))
  fam.vals <- dat[, c(CPD$parents, CPD$id), drop=FALSE]
  print(fam.vals)
  np <- length(CPD$parents)
  if (np == 1)
    CPD$tr <- tree(V2 ~ ., data = fam.vals)
  else if (np == 2)
    CPD$tr <- tree(V3 ~ ., fam.vals)
  else if (np == 3)
    CPD$tr <- tree(V4 ~ ., fam.vals)
  else
    stop(paste("too many parents for tree CPD", CPD$id))
  end
  list(CPD = CPD, dev = deviance(CPD$tr), score = dev)
}


Thanks for your help!

Kevin

  
P.S. You can reproduce the above problem using the following script

Nvars <- 4;
Nsamples <- 10;
N <- Nvars*Nsamples;
dat <- matrix(sample(1:2, N, c(0.5,0.5), replace=TRUE), Nsamples, Nvars)
#dat <- sample.bnet(bnet, 10)
dat <- as.data.frame(dat)
for (i in 1:Nvars) {
   dat[,i]<- factor(dat[,i], 1:2)
}
tr <- tree(V4 ~ ., dat)
#deviance(tr)

foo <- function(dummy) {
  tr <- tree(V4 ~ ., dummy)
}
tr2 <- foo(dat)
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Peter Dalgaard
# 
Kevin Murphy <murphyk at cs.berkeley.edu> writes:
Jonathan Li reported the same thing just three days ago on R-help. It
happens because tree uses the S-PLUS form of passing the frame number
of the parent (i.e. eval(model, sys.parent())) rather than the parent
environment, which conspires with an age-old bug in R that makes this
work only on the command line. The quick fix is to modify the tree
function to use eval(model, parent.frame()) instead. This is better R
style anyway, but we'll get around to fixing eval() eventually.