melt array with multiple value variable - r

Question

I have a multidimensional array a, and I want to format it into out. I used melt followed by dcast, but I wonder if there is a better way of doing it, without or without using library(reshape)?
library(reshape2)
(a=array(1:3^4,c(3,3,3,3),dimnames=list("d1"=paste("d1",letters[1:3],sep="-"),
"d2"=paste("d2",letters[1:3],sep="-"),
"d3"=paste("d3",letters[1:3],sep="-"),
"d4"=paste("d4",letters[1:3],sep="-"))))
(out=dcast(melt(a,id.vars=c("d1","d2","d3")),d1+d2+d3~d4))
I am asking this question because
My solution feels somewhat repetitive because I am using melt followed by cast and specifying d1,d2,d3 two times. I wonder if there is a more straightforward way of doing things.
It would be a good if there is an at least as compact solution that doesn't require loading an extra package.
So to reiterate, I will be happy with any of the following:
A more straightforward solution that requires reshape2
A more straightforward solution that doesn't require reshape2
An at least as compact solution that doesn't require reshape2

Answer 1

I'm assuming you won't just be copying and pasting code, but rather, either sharing a collection of scripts that could be sourced, or even creating a package of your functions.
Keeping that in mind, it's easy for you to recreate the function that I referred to in the comments.
Here's ftable(a):
ftable(a)
# d4 d4-a d4-b d4-c
# d1 d2 d3
# d1-a d2-a d3-a 1 28 55
# d3-b 10 37 64
# d3-c 19 46 73
# d2-b d3-a 4 31 58
# d3-b 13 40 67
# d3-c 22 49 76
# d2-c ......................
# ................................
And its attributes:
attributes(ftable(a))
# $dim
# [1] 27 3
#
# $class
# [1] "ftable"
#
# $row.vars
# $row.vars$d1
# [1] "d1-a" "d1-b" "d1-c"
#
# $row.vars$d2
# [1] "d2-a" "d2-b" "d2-c"
#
# $row.vars$d3
# [1] "d3-a" "d3-b" "d3-c"
#
#
# $col.vars
# $col.vars$d4
# [1] "d4-a" "d4-b" "d4-c"
You can use these attributes to create a function that looks like this:
ftable2df <- function (mydata) {
if (class(mydata) != "ftable") mydata <- ftable(mydata)
dfrows <- rev(expand.grid(rev(attr(mydata, "row.vars"))))
dfcols <- as.data.frame.matrix(mydata)
names(dfcols) <- do.call(
paste, c(rev(expand.grid(rev(attr(mydata, "col.vars")))),
sep = "_"))
cbind(dfrows, dfcols)
}
ftable2df(a)
# d1 d2 d3 d4-a d4-b d4-c
# 1 d1-a d2-a d3-a 1 28 55
# 2 d1-a d2-a d3-b 10 37 64
# 3 d1-a d2-a d3-c 19 46 73
# 4 d1-a d2-b d3-a 4 31 58
# 5 d1-a d2-b d3-b 13 40 67
# 6 d1-a d2-b d3-c 22 49 76
# 7 d1-a d2-c d3-a 7 34 61
# 8 d1-a d2-c d3-b 16 43 70
# 9 d1-a d2-c d3-c 25 52 79
# 10 d1-b d2-a d3-a 2 29 56
# 11 d1-b d2-a d3-b 11 38 65
# 12 d1-b d2-a d3-c ............
# ................................
Update (non-base solution)
If you're not married to "reshape2" and are open to using a package as long as it's on CRAN, and if you are open to a solution that might be a little slower than melting and dcasting your data, you can also look at adply from "plyr".
library(plyr)
adply(a, 1:3)

Answer 2

An alternative that 1) is short 2) only uses base R
cbind(do.call(expand.grid, dimnames(a)[1:3]), apply(a, 4, identity))
# d1 d2 d3 d4-a d4-b d4-c
#1 d1-a d2-a d3-a 1 28 55
#2 d1-b d2-a d3-a 2 29 56
#3 d1-c d2-a d3-a 3 30 57
# etc
My original solution used reshape and was a bit goofy... I think this is preferable by a long way.