Question

I am wondering why the following bit of code doesn’t produce the two plots one below the other.

data(mtcars)
library(randomForest)

mtcars.rf <- randomForest(mpg ~ ., data=mtcars, ntree=1000, keep.forest=FALSE,
                          importance=TRUE)
png("rf1.png", width=6, height=6, units="in", res=100)
par(mfcol=c(2,1))
varImpPlot(mtcars.rf)
plot(mtcars.rf, log="y")
dev.off()

This just produces

with a blank second row.

Answer 1

The problem is that varImpPlot redefines the plotting area, and then resets it to the previous value. This means it acts as if you called par(mfcol=c(2,1)) after the varImpPlot line. If you extract the varImpPlot plotting data you can plot the two dotcharts yourself (you can use layout rather than par to split the plotting area into different shaped regions):

data(mtcars)
library(randomForest)

mtcars.rf <- randomForest(mpg ~ ., data=mtcars, ntree=1000, keep.forest=FALSE,
                          importance=TRUE)
varImpData <- varImpPlot(mtcars.rf) # calculate this outside the plot

png("rf1.png", width=6, height=6, units="in", res=100)
layout(matrix(c(1,2,3,3), 2, 2, byrow = TRUE))
dotchart(varImpData[,c(1)])
dotchart(varImpData[,c(2)])
plot(mtcars.rf, log="y")
dev.off()