named_list_timing.R

## l1 is a list with 20 numeric elements of length 1e4 each

set.seed(123)
l2 <- l1 <- replicate(20,rnorm(1e4),simplify=FALSE)
length(l1)
sapply(l1,length)

## l2 is the same list with names
names(l2) <- paste0("X", 1:length(l2))

all.equal(l1, l2)
## [1] "names for current but not for target"

library("rbenchmark")
benchmark(range(l1),
          range(l2),
          range(sapply(l2, range)))[, 1:5]

##                       test replications elapsed relative user.self
## 1                range(l1)          100   0.992    1.023     0.988
## 2                range(l2)          100  31.703   32.684    31.548
## 3 range(sapply(l2, range))          100   0.970    1.000     0.968

## Rprof indicates that it was the c primitive that accounts for the
## big different in time.

## > range.default
## function (..., na.rm = FALSE, finite = FALSE)
## {
##     x <- c(..., recursive = TRUE)
##     if (is.numeric(x)) {
##         if (finite)
##             x <- x[is.finite(x)]
##         else if (na.rm)
##             x <- x[!is.na(x)]
##         return(c(min(x), max(x)))
##     }
##     c(min(x, na.rm = na.rm), max(x, na.rm = na.rm))
## }

## The line that recursively concatenates the ... arguments transforms
## the list in a numeric and names all individual elements, thus
## generating a _named_ numeric of length 20 times 10000. As
## demonstrated below, it is the creation of the names that accounts
## for the overhead.

benchmark(c(l2, recursive = TRUE),
          c(l1, recursive = TRUE))[, 1:5]

##                      test replications elapsed relative user.self
## 2 c(l1, recursive = TRUE)          100   0.258    1.000     0.256
## 1 c(l2, recursive = TRUE)          100  27.879  108.058    27.776