Question

I’m trying to create a hard-copy image of a neural network graph and it keeps failing. If I try to create a PNG, nothing is generated, and if I try to generate a PDF I get a small file output that refuses to open with “file may be damaged” errors. If I just let it display in a graphics window, the image comes up fine.

I’m using 2.15.1 on OS X (10.7.4), built by Macports. The code I’m working with at the moment:

library(ALL)
library(neuralnet)
data(ALL)

ALL.pdat <- pData(ALL)
bt <- factor(substring(ALL.pdat$BT,1,1))
all.sds <- apply(exprs(ALL),1,sd)
top.10.sds <- rank(all.sds)>length(all.sds)-10
exprs.top.10 <- as.data.frame(t(exprs(ALL)[top.10.sds,]))
nn.data <- cbind(exprs.top.10, as.numeric(bt))
##  Gene names start with a number, and that causes problems when trying to set up the
##  formula for neuralnet.
col.names <- paste("g", colnames(nn.data), sep = '')
col.names[11] <- "bt"
colnames(nn.data) <- col.names
my.nn <- neuralnet(bt ~ g36108_at + g36638_at + g37006_at + g38096_f_at + g38319_at + g38355_at + g38514_at + g38585_at + g39318_at + g41214_at, nn.data,  hidden = 10, threshold = 0.01)
summary(my.nn)
pdf("./nn-all.pdf")
plot.nn(my.nn)
dev.off()
png("./nn-all.png")
plot.nn(my.nn)
dev.off()

I’ve even rebooted the machine to make sure that all the memory is cleared up, and that didn’t help any.

Answer 1

Simple reproducible example:

pdf("test.pdf")
 set.seed(42)
 plot(runif(20),rnorm(20))
 png("test.png")
  set.seed(42)
  plot(runif(20),rnorm(20))
 dev.off()

If I try to open the PDF with Adobe Reader on my German Windows 7, I get a nice informative error message telling me that the file cannot be opened because the file is in use by another application. This can be fixed easily:

pdf("test.pdf")
 set.seed(42)
 plot(runif(20),rnorm(20))
dev.off() #make sure to close the graphics device
png("test.png")
 set.seed(42)
 plot(runif(20),rnorm(20))
dev.off()

Edit:

The problem is plot.nn. Until the package gets patched, you need to redefine plot.nn manually as shown in this answer.