less clunky reshaping of anscombe data - r

Question

I was trying to use ggplot2 to plot the built-in anscombe data set in R (which contains four different small data sets with identical correlations but radically different relationships between X and Y). My attempts to reshape the data properly were all pretty ugly. I used a combination of reshape2 and base R; a Hadleyverse 2 (tidyr/dplyr) or a data.table solution would be fine with me, but the ideal solution would be
short/no repeated code
comprehensible (somewhat conflicting with criterion #1)
involve as little hard-coding of column numbers, etc. as possible
The original format:
anscombe
## x1 x2 x3 x4 y1 y2 y3 y4
## 1 10 10 10 8 8.04 9.14 7.46 6.58
## 2 8 8 8 8 6.95 8.14 6.77 5.76
## 3 13 13 13 8 7.58 8.74 12.74 7.71
## ...
## 11 5 5 5 8 5.68 4.74 5.73 6.89
Desired format:
## s x y
## 1 1 10 8.04
## 2 1 8 6.95
## ...
## 44 4 8 6.89
Here's my attempt:
library("reshape2")
ff <- function(x,v)
setNames(transform(
melt(as.matrix(x)),
v1=substr(Var2,1,1),
v2=substr(Var2,2,2))[,c(3,5)],
c(v,"s"))
f1 <- ff(anscombe[,1:4],"x")
f2 <- ff(anscombe[,5:8],"y")
f12 <- cbind(f1,f2)[,c("s","x","y")]
Now plot:
library("ggplot2"); theme_set(theme_classic())
th_clean <-
theme(panel.margin=grid::unit(0,"lines"),
axis.ticks.x=element_blank(),
axis.text.x=element_blank(),
axis.ticks.y=element_blank(),
axis.text.y=element_blank()
)
ggplot(f12,aes(x,y))+geom_point()+
facet_wrap(~s)+labs(x="",y="")+
th_clean

Answer 1

If you are really dealing with the "anscombe" dataset, then I would say #Thela's reshape solution is very direct.
However, here are a few other options to consider:
Option 1: Base R
You can write your own "reshape" function, perhaps something like this:
myReshape <- function(indf = anscombe, stubs = c("x", "y")) {
temp <- sapply(stubs, function(x) {
unlist(indf[grep(x, names(indf))], use.names = FALSE)
})
s <- rep(seq_along(grep(stubs[1], names(indf))), each = nrow(indf))
data.frame(s, temp)
}
Notes:
I'm not sure that this is necessarily less clunky than what you're already doing
This approach will not work if the data are "unbalanced" (for example, more "x" columns than "y" columns.)
Option 2: "dplyr" + "tidyr"
Since pipes are the rage these days, you can also try:
library(dplyr)
library(tidyr)
anscombe %>%
gather(var, val, everything()) %>%
extract(var, into = c("variable", "s"), "(.)(.)") %>%
group_by(variable, s) %>%
mutate(ind = sequence(n())) %>%
spread(variable, val)
Notes:
I'm not sure that this is necessarily less clunky than what you're already doing, but some people like the pipe approach.
This approach should be able to handle unbalanced data.
Option 3: "splitstackshape"
Before #Arun went and did all that fantastic work on melt.data.table, I had written merged.stack in my "splitstackshape" package. With that, the approach would be:
library(splitstackshape)
setnames(
merged.stack(
data.table(anscombe, keep.rownames = TRUE),
var.stubs = c("x", "y"), sep = "var.stubs"),
".time_1", "s")[]
A few notes:
merged.stack needs something to treat as an "id", hence the need for data.table(anscombe, keep.rownames = TRUE), which adds a column named "rn" with the row numbers
The sep = "var.stubs" basically means that we don't really have a separator variable, so we'll just strip out the stub and use whatever remains for the "time" variable
merged.stack will work if the data are unbalanced. For instance, try using it with anscombe2 <- anscombe[1:7] as your dataset instead of "anscombe".
The same package also has a function called Reshape that builds upon reshape to let it reshape unbalanced data. But it's slower and less flexible than merged.stack. The basic approach would be Reshape(data.table(anscombe, keep.rownames = TRUE), var.stubs = c("x", "y"), sep = "") and then rename the "time" variable using setnames.
Option 4: melt.data.table
This was mentioned in the comments above, but hasn't been shared as an answer. Outside of base R's reshape, this is a very direct approach that handles column renaming from within the function itself:
library(data.table)
melt(as.data.table(anscombe),
measure.vars = patterns(c("x", "y")),
value.name=c('x', 'y'),
variable.name = "s")
Notes:
Will be insanely fast.
Much better supported than "splitstackshape" or reshape ;-)
Handles unbalanced data just fine.

Answer 2

I think this meets the criteria of being 1) short 2) comprehensible and 3) no hardcoded column numbers. And it doesn't require any other packages.
reshape(anscombe, varying=TRUE, sep="", direction="long", timevar="s")
# s x y id
#1.1 1 10 8.04 1
#...
#11.1 1 5 5.68 11
#1.2 2 10 9.14 1
#...
#11.2 2 5 4.74 11
#1.3 3 10 7.46 1
#...
#11.3 3 5 5.73 11
#1.4 4 8 6.58 1
#...
#11.4 4 8 6.89 11

Answer 3

I don't know if a non-reshape solution would be acceptable, but here you go:
library(data.table)
#create the pattern that will have the Xs
#this will make it easy to create the Ys
pattern <- 1:4
#use Map to create a list of data.frames with the needed columns
#and also use rbindlist to rbind the list produced by Map
lists <- rbindlist(Map(data.frame,
pattern,
anscombe[pattern],
anscombe[pattern+length(pattern)]
)
)
#set the correct names
setnames(lists, names(lists), c('s','x','y'))
Output:
> lists
s x y
1: 1 10 8.04
2: 1 8 6.95
3: 1 13 7.58
4: 1 9 8.81
5: 1 11 8.33
6: 1 14 9.96
7: 1 6 7.24
8: 1 4 4.26
9: 1 12 10.84
10: 1 7 4.82
....

Answer 4

A newer tidyverse option is suggested in the tidyverse vignette:
anscombe %>%
pivot_longer(everything(),
names_to = c(".value", "set"),
names_pattern = "(.)(.)"
) %>%
arrange(set)
#> # A tibble: 44 x 3
#> set x y
#> <chr> <dbl> <dbl>
#> 1 1 10 8.04
#> 2 1 8 6.95
#> 3 1 13 7.58
#> 4 1 9 8.81
#> 5 1 11 8.33
#> 6 1 14 9.96
#> 7 1 6 7.24
#> 8 1 4 4.26
#> 9 1 12 10.8
#> 10 1 7 4.82
#> # … with 34 more rows