R help - Feb 2013 - Parallelizing Other Apply Functions, e.g. by, the Easy (Wrong?) Way

Dear R Users---this is more curiosity than a real problem.  I am wondering
how to add mc* functions for all of R's *apply functions.   stackoverflow
3505701 has a nice overview of these functions.  roughly,

apply ( function to rows and columns of matrix )
lapply ( function to each element of list, get back list )
sapply ( function to each element of list, get back vector )
vapply ( like sapply, but tells R about return values of function for speed
)
mapply ( function to first element of multiple lists, then second element
of multiple lists, etc )
rapply ( uncommon recursive use )

tapply ( function to one vector based on groups of another vector )
by ( function to apply to data frame or list, based on one vector in the
list; relies on tapply )
aggregate ( seems to be sort of like "by" )

ave ( merges by results back into a vector of same length )


many of the functions use lapply internally.  "by" uses tapply
internally.
which uses lapply.  these functions should ideally have had a parameter
that is the lapply function that they use, so that mc.by <- function(...)
by(..., lapply=mclapply ).  unfortunately, they do not.  so, I am now
wondering what the preferred way is to "patch" them.

one way to do this would be:

  mc.by <- function( ... ) {
      oc.lapply <- lapply
      lapply <<- mclapply
      result <- by( ... )
      lapply <<- oc.lapply
      return(result)
  }


the disadvantage is that if the function to mc.by itself relies on, say,
sapply somewhere, both would use the multicore mclapply function.  I don't
think it is possible to spoof the GlobalEnv for only one function, but not
its own lower-tier functions.  another big problem here is that I probably
have to trap this function appropriately, so that this restores on abort
the original lapply function.  not pretty.  the advantage of doing it is
that future changes by the R core team to by() will not create any changes.

the alternative is to copy the definitions from by.data.frame and
by.default, and replace them with my own, which has only one change---the
optional argument.  this is not hard to do, but I now run the risk that the
R team could change by().  I wish I could at least test whether the by()
function changes from release to release to alert me, but functions are not
atomic and therefore cannot be compared.


what is the recommended way to do this?

/iaw

----
Ivo Welch (ivo.welch@gmail.com)
http://www.ivo-welch.info/

	[[alternative HTML version deleted]]