Instead of the normal R data frame, you can use a immutable data frame which returns pointers to the original when you subset and can be much faster:
idf <- idata.frame(myDF)
system.time(aggregateDF <- ddply(idf, c("year", "state", "group1", "group2"),
function(df) wtd.mean(df$myFact, weights=df$weights)))
# user system elapsed
# 18.032 0.416 19.250
If I was to write a plyr function customised exactly to this situation, I’d do something like this:
system.time({
ids <- id(myDF[c("year", "state", "group1", "group2")], drop = TRUE)
data <- as.matrix(myDF[c("myFact", "weights")])
indices <- plyr:::split_indices(seq_len(nrow(data)), ids, n = attr(ids, "n"))
fun <- function(rows) {
weighted.mean(data[rows, 1], data[rows, 2])
}
values <- vapply(indices, fun, numeric(1))
labels <- myDF[match(seq_len(attr(ids, "n")), ids),
c("year", "state", "group1", "group2")]
aggregateDF <- cbind(labels, values)
})
# user system elapsed
# 2.04 0.29 2.33
It’s so much faster because it avoids copying the data, only extracting the subset needed for each computation when it’s computed. Switching the data to matrix form gives another speed boost because matrix subsetting is much faster than data frame subsetting.