purrr_example3.R

nrep <- 100
d <- expand.grid(n=c(3,5,10,20), type=c("normal","t"))
d <- d[rep(seq_len(nrow(d)),each=nrep),]

simulate <- function(n, type) {
  if (type == "normal") {
    dat <- rnorm(n)
  } else {
      dat <- rt(n, df=3)
    }
  dat
  meanerr <- abs(mean(dat) - 0)
  medianerr <- abs(median(dat) - 0)
  c(mean=meanerr, median=medianerr)
}

library(purrr)
library(dplyr)

# for each row, take columns 'n' and 'type' and feed these
# to simulate(), then stash the result as a list in the 'errs' 
# column. then i unpack the list as two columns of doubles
# (maybe there's a better way to do unpacking)
# but note that you can stash anything in a cell in the data frame, 
# even complex data objects like data frames.
d2 <- d %>% mutate(
  errs = map2(n, type, simulate),
  mean = map_dbl(errs, `[`, 1),
  median = map_dbl(errs, `[`, 2)
  )

d2 <- select(d2, -errs)
d2$n <- factor(d2$n)

library(reshape)
m <- melt(d2, id.vars=c("n","type"))

library(ggplot2)
ggplot(m, aes(x=n, y=value, col=variable)) + geom_boxplot() +
  facet_wrap(~ type) + ylim(0,2)

this is probably not the best purrr example. note that you can put anything in the data frame, here i just store two numbers, but you can put a data object as a cell in the data frame, e.g. another data frame.

for more see blogpost:
http://blog.rstudio.org/2015/09/29/purrr-0-1-0/

or pkg description:
https://github.com/hadley/purrr/blob/master/README.md