ggplot2 multiplot using changing variables - r

Question

I am trying to create multiple plots using ggplot2 that is then gathered in using multiplot. However, when I try to create X graphs I end up with X of the same graph.
My problem code pretty much boils down to this, asuming df is the dataframe
library(ggplot2)
i = 1
j = 2
xVar = df[[i]]
yVar = df[[j]]
plot1 = ggplot(data = df, aes(xVar, yVar)) + geom_point(shape=1)
i = 1
j = 3
xVar = df[[i]]
yVar = df[[j]]
plot2 = ggplot(data = df, aes(xVar, yVar)) + geom_point(shape=1)
multiplot(plot1,plot2, cols=2)
At this point plot1 is equal to plot2 and I dont understand why.
My full code if interested:
n = 1
columns = colnames(df)
plots = list()
for(i in 3:7)
{
for(j in (i+1):7)
{
if(j < 8 & i < 7) {
xVar = df[[i]]
yVar = df[[j]]
plots[[n]] = ggplot(data = df, aes(x=xVar, y=yVar)) +
geom_point(shape=1) +
labs(x=columns[[i]], y=columns[[j]]) +
theme(axis.title=element_text(size=8))
n = n + 1
}
}
}
multiplot(plotlist = plots, cols=3)

Answer 1

There are lots of things going on here.
First, it is a really, really, really bad idea to use external variables in calls to aes(...). The arguments to aes(...) are evaluated in the context of the data=... argument, so in the context of df in your case. If that fails they are evaluated in the global environment. So it is highly preferable to do something like this:
gg <- data.frame(x=df[[i]],y=df[[j]])
plots[[n]] = ggplot(data = gg, aes(x,y)) +...
Second, ggplot stores the expressions from aes(...) and evaluates them when the plot is rendered (so, during the call to multiplot(...)). All of your plots use variables named xVar and yVar in aes(...). So when these plots are rendered, ggplot uses whatever is stored in those variables at the time - presumably from the last plot definition. That's why all your plots look like the last one. This is the reference to "lazy evaluation" in the other answer.
On the other hand, ggplot evaluates the data=... argument immediately, and stores the dataset as part of the plot definition (in the gtable). So creating different data frames (called gg above), for each plot will work.
Finally, it looks like you are trying to create a pairs plot (every column vs. every other column, more or less). Unless this is a homework assignment, there are much easier ways to do this. You could use ggpairs(...) in the GGally package (which uses grid graphics), or you could do it this way using basic ggplot with facets:
# make up some data
set.seed(1) # for reproducible example
df <- data.frame(matrix(rnorm(700),nc=7))
df[4] <- 1+2*df[3] + rnorm(100)
df[5] <- 3*df[3] - 2*df[4] + rnorm(100)
df[6] <- -10*df[5] + rnorm(100)
# you start here...
gg.pairs <- function(data) { # scatterplot matrix using ggplot facets
require(ggplot2)
require(data.table)
require(reshape2) # for melt(...)
DT <- data.table(melt(cbind(id=1:nrow(data),data),id="id"),key="id")
gg <- DT[DT,allow.cartesian=T]
setnames(gg,c("id","H","x","V","y"))
ggplot(gg[as.integer(gg$H)<as.integer(gg$V),], aes(x,y)) +
geom_point(shape=1) +
facet_grid(V~H, scales="free")
}
gg.pairs(df[3:7])

Answer 2

I think that your problem is in R lazy evaluation. Indeed what happens is that plot1 and plot2 are not created when you assign it but when you call it, and at this moment there is only one copy (the last one) of xVarand yVar and plots are the same

Answer 3

Well, I can't explain what is happening, but a workaround is to use column names instead of columns withaes_string. The following makes two unique plots in multiplot for me, and this change could easily be incorporated into your plot loop.
dat = data.frame(x = rnorm(10), y1 = rnorm(10), y2 = rpois(10, 5))
xVar = names(dat)[1]
yVar = names(dat)[2]
plot1 = ggplot(data = dat, aes_string(xVar, yVar)) + geom_point(shape=1)
yVar = names(dat)[3]
plot2 = ggplot(data = dat, aes_string(xVar, yVar)) + geom_point(shape=1)
multiplot(plot1, plot2, cols=2)