问题
I have a for loop which produces 60 plots. I would like to save all this plots in only one file.
If I set par(mfrow=c(10,6)) it says : Error in plot.new() : figure margins too large
What can I do?
My code is as follows:
pdf(file="figure.pdf")
par(mfrow=c(10,6))
for(i in 1:60){
x=rnorm(100)
y=rnorm(100)
plot(x,y)
}
dev.off()
回答1:
Your default plot, as stated in the loop, does not use the space very effectively. If you look at just a single plot, you can see it has large margins, both between axis and edge and plot area and axis text. Effectively, there is a lot of space-hogging.
Secondly, the default pdf-function creates small pages, 7 by 7 inches. That is not a large sheet to plot on.
Trying to plot a 10 x 6 or 12 x 5 on 7 by 7 inches is therefore trying to squeeze in a lot of whitespace on very little space.
For it to succeed, you must look into the margin-options of par which is mar, mai, oma and omi, and probably some more. Consult the documentation with the command
?par
In addition to this, you could consider not displaying axis-text, tick-marks, tick-labels and titles for every one of the 60 sub-plots, as this too will save you space.
But somebody has already gone through some of this trouble for you. Look into the lattice-package or ggplot2, which has some excellent methods for making table-like subplots.
But there is another pressing issue: What are you trying to display with 60 subplots?
Update
Seeing what you are trying to do, here is a small example of faceting in ggplot2. It uses the Tufte-theme from jrnold's ggthemes, which is copied into here and then modified slightly in the line after the function.
library(ggplot2)
library(scales)
#### Setup the `theme` for the plot, i.e. the appearance of background, lines, margins, etc. of the plot.
## This function returns a theme-object, which ggplot2 uses to control the appearance.
theme_tufte <- function(ticks=TRUE, base_family="serif", base_size=11) {
ret <- theme_bw(base_family=base_family, base_size=base_size) +
theme(
legend.background = element_blank(),
legend.key = element_blank(),
panel.background = element_blank(),
panel.border = element_blank(),
strip.background = element_blank(),
plot.background = element_blank(),
axis.line = element_blank(),
panel.grid = element_blank())
if (!ticks) {
ret <- ret + theme(axis.ticks = element_blank())
}
ret
}
## Here I modify the theme returned from the function,
theme <- theme_tufte() + theme(panel.margin=unit(c(0,0,0,0), 'lines'), panel.border=element_rect(colour='grey', fill=NA))
## and instruct ggplot2 to use this theme as default.
theme_set(theme)
#### Some data generation.
size = 60*30
data <- data.frame(x=runif(size), y=rexp(size)+rnorm(size), mdl=sample(60,size, replace=TRUE))
#### Main plotting routine.
ggplot(data, aes(x,y, group=mdl)) ## base state of the plot to be used on all "layers", i.e. which data to use and which mappings to use (x should use x-variable, y should use the y-variable
+ geom_point() ## a layer that renders data as points, creates the scatterplot
+ stat_quantile(formula=y~x) ## another layer that adds some statistics, in this case the 25%, 50% and 75% quantile lines.
+ facet_wrap(~ mdl, ncol=6) ## Without this, all the groups would be displayed in one large plot; this breaks it up according to the `mdl`-variable.
The usual challenge in using ggplot2 is restructuring all your data into data.frames. For this task, the reshape2 and plyr-packages might be of good use.
For you, I would imagine that your function that creates the subplot both calculates the estimation and creates the plot. This means that you have to split the function into calculating the estimation, returning it to a data.frame, which you then can collate and pass to ggplot.
回答2:
Output the plots to a pdf:
X = matrix(rnorm(60*100), ncol=60)
Y = matrix(rnorm(60*100), ncol=60)
pdf(file="fileName.pdf")
for(j in 1:60){
plot(X[,j], Y[,j])
}
dev.off()
回答3:
For placing many plots on a page or document (and I have created images with literally thousands of plots in them), it is convenient to separate the work between R--which creates the plots individually--and other software which is better suited for arranging arrays of things. If this reminds you of spreadsheets or word processing tables, then we are thinking alike.
This page, which is a screenshot from a PDF file, contains over 200 statistical graphics. Although it has been greatly reduced (to 40% nominal size) in order to obscure proprietary data, the original has all the detail of the original R graphics and can be zoomed to 1600% without problem.
Two mechanisms have worked reasonably well. For up to several hundred plots, a little macro to import and re-sequence a set of bitmapped image files (.emf or .wmf) into a Word document does fine. For better control, I turn to a comparable Excel macro. It is driven by a sheet that is empty of everything except a row with column headers and a column with row headers. (You can see them at the left and top of the figure.) The macro deletes everything else on that sheet (except for formatting), then munges each possible combination of row and column header into a file name and if it finds that file, it imports it into the corresponding cell. The whole operation takes just a few seconds for several thousand images.
Obviously this communication mechanism between R and the other software is primitive, consisting of a collection of image files having a standard naming convention. But the code needed to implement it all is brief (albeit customized to each situation) and it works reliably. For example, if you encapsulate the plotting code within a function, then it will be called within a loop to create many similar plots. At the end of that function add a few lines to save the plot to a file, something like this:
path <- "W: <whatever>/" # Folder for the output files
ext <- "wmf" # or "emf" or "png" or ... # Format (and extension) of the output
...
if (save) {
outfile <- paste(path, paste(munge(well), munge(parm), sep="_"), sep="/")
outfile <- paste(outfile, ext, sep=".")
savePlot(filename=outfile, type=ext)
}
In this case each plot is identified by two loop variables, well and parm, both of which are strings (they correspond to the column and row headers). The function for creating acceptable filenames merely strips out punctuation, replacing it by an anodyne placeholder:
munge <- function(s) gsub("[[:punct:]]", "_", s)
Once those images have been imported into Word, Excel, or wherever you like, it's fairly easy to reorganize them, place other material around them, etc., and then print the result in PDF format.
There is an art to creating these very large "small multiples" (in Tufte's terminology). To the extent possible, it helps to follow Tufte's principle of increasing the data:ink ratio by erasing inessential material. That makes graphical patterns clear even when the tableau has been greatly reduced in size in order to comprehend all its rows and columns at once. Although the preceding figure is a poor example--the individual plots had to have axes, gridlines, labels, and so on so that they can be read in detail when zoomed--the power of this method to reveal patterns is clear even at this scale. It is crucial to make the plots comparable to one another. In this example, which consists of time series, every plot has the same range on the x-axis; within each row (which corresponds to a different type of observation), the ranges on the y-axes are the same; and all color schemes and methods of symbolization are the same throughout.
回答4:
You could also use knitr. This didn't instantly convert over to base graphics (and I've got to run now), but using ggplot works easily.
\documentclass{article}
\begin{document}
<<echo = FALSE, fig.keep='high', fig.height=3, fig.width=4>>=
require(ggplot2)
for (i in 1:10) print(ggplot(mtcars, aes(x = disp, y = mpg)) + geom_point())
@
\end{document}
The above code will produce a nice multi-page pdf with all the graphs.
回答5:
For a very simple solution to this type of issue, I found that setting a large "Windows" device manages to make the window big enough for many uses.
windows(50,50)
par(mfrow=c(10,6))
for(i in 1:60){
x=rnorm(100)
y=rnorm(100)
plot(x,y)
}
Or in my case,
windows(20,20)
plot(Plotting_I_Need_In_Rows_of_4, mfrow=c(4,4))
来源：https://stackoverflow.com/questions/16698741/how-to-deal-with-a-lot-of-plots-in-r