How to join efficiently multiple rgl plots into one single plot?

Question

I produced 3D plots with the rgl package for each factor level of my data and saved them as pngs. My data has 30 different levels, which resulted in 30 different image files. Now I would like to combine these pngs into one single plot.

I would display them like this :

enter image description here

The following example illustrates what I would like to do:

library(rgl)
library(png)
library(gridExtra)
library(ggplot2)

## creates a png in the working directory which can be used as an example
example(surface3d)
rgl.snapshot("example.png")
rgl.close()

## imports the png files; in the example, the same file is imported multiple times.
if(exists("png.df")) rm(png.df)
for (i in 1:9) {
  png.i <- readPNG("example.png")

  g <- rasterGrob(png.i, interpolate=TRUE)
  g <- g$raster
  g <- as.vector(g)
  g <- matrix(g, nrow = 256, ncol = 256, dimnames = list(1:256, 1:256))

  df.i <- data.frame(i = rep(row.names(g), dim(g)[2]), j = rep(colnames(g), each = dim(g)[1]), col=as.vector(g))
  df.i$i <- as.numeric(as.character(df.i$i))
  df.i$j <- as.numeric(as.character(df.i$j))
  df.i$col <- as.character(df.i$col)
  df.i$title <- paste ( "Plot", i)

  if(exists("png.df")) {
    png.df <- rbind(png.df, df.i)
  } else {
    png.df <- df.i
  }
}
rm(df.i, g)

## plots the data
pl <- ggplot(png.df, aes( x = i, y = j))
pl <- pl + geom_raster(aes(fill = col)) + scale_fill_identity()
pl <- pl + scale_y_reverse()
pl <- pl + facet_wrap( ~ title)
pl <- pl + coord_equal() + theme_bw() + theme(panel.grid = element_blank(), axis.text = element_blank(), axis.title = element_blank(), axis.ticks= element_blank())
pl

This works rather well, but it is quite slow. The real pngs have a much higher resolution, and I'd like to plot 30 pngs, not just 9, which results in my machine being totally unresponsive for quite a long time (i7, 8GB RAM).

The importation part works reasonably well, but the resulting data frame is extremely big (4.5e+07 rows), which ggplot (understandably) can't handle properly.

How could a plot be created in a fast and efficient manner? Preferably with R, but other software might also be used.

How you will plot 30 png in the same plot? you will overlay them? it like you have 30 matrix and you want to display them, so you will mean them? or stack them? — agstudy, Mar 05 '13 at 11:15
They should be displayed in a grid, just like [this](http://docs.ggplot2.org/0.9.3/facet_wrap-2.png). — Jon Snow, Mar 05 '13 at 11:31
A warning: don't trust your friends. They may have taken the black and elected you their leader but they'll turn on you in an instant. (OK, sorry, but you asked for that when you took that name) — Carl Witthoft, Mar 05 '13 at 12:12
Might be worth experimenting with reduced resolution to see whether the final output (30 small charts on one page) looks any different. I'm betting that the fastest approach will be to recreate the 30 charts at a much lower resolution and pumping these files into `ggplot` or `grid` to create the combined layout. — Carl Witthoft, Mar 05 '13 at 14:52
@JonSnow i think 30 pictures is a lot ..maybe it is better to divide it by 2.. — agstudy, Mar 05 '13 at 16:49

score 8 · Accepted Answer · edited Mar 05 '13 at 19:42

8

Here's a solution using the grid functiongrid.raster and xyplot from lattice. I think that grid.raster has faster rendering to screen, so it is a good candidate for performance. I am choosing lattice because it integrate easier the grid function using the panel customization.

enter image description here

First I read all the png, using readPNG from png package ( similar to your solution)

ll <- list.files(path='c:/temp',patt='compo[0-9].*',full.names=T)
library(png)
imgs <- lapply(ll,function(x){
       as.raster(readPNG(x))  ## no need to convert to a matrix here!
   })

Then I prepare data for scatter plot:

x = 1:4   ## here 4 because I use  16 plots
y = 1:4
dat <- expand.grid(x,y)

Finaly I use xyplot with a custom panel function:

library(lattice)
library(grid)
xyplot(Var2~Var1|rownames(dat),data=dat,layout=c(4,4),
      panel=function(x,y,...){
        lims <- current.panel.limits()
        grid.raster(image =imgs[[panel.number()]],sum(lims$xlim)/2,sum(lims$ylim)/2,
                                      width =diff(lims$xlim),
                                          height=diff(lims$ylim),def='native' )

       })

PS: it is what I call a catty solution.

edited Mar 05 '13 at 19:42

Josh O'Brien

159,210
26
366
455

answered Mar 05 '13 at 16:48

agstudy

119,832
17
199
261

This solution runs like a cat on catnip! It's at least 10 times faster than the `ggplot` approach, and much more elegant. Thanks! – Jon Snow Mar 06 '13 at 08:44
@JonSnow did you test it over 32 png or more? the solution as it is not finished, if you see the panels are not in the good order, maybe you will need to fix that. – agstudy Mar 06 '13 at 08:47
@agstudy Yes, with 32 pngs (each 1500x1000), it takes around 10 secs to import, and less than 40 secs to create the plot. The order of the panels is indeed not correct. I've never worked with `lattice` before, but this would be a good occasion to start. – Jon Snow Mar 06 '13 at 08:57
@JonSnow here it is fast because the grid.raster is fast ( it is created to manipulate images) , lattice used for high levels functionalities : scales, panels... – agstudy Mar 06 '13 at 09:00

How to join efficiently multiple rgl plots into one single plot?

1 Answers1

Linked