Related
In the R plotting environment, is there a way to specify a universal distance between the placement of the axis label and axis name relative the the boundary of the plot? For example, consider the Y-axis of the following test plot:
pdf('test1.pdf',height=3.5,width=5.0)
plot(1,ylab=NA,yaxt='n',ann=F)
mtext('Y-Axis Label',side=2,line=1.5)
axis(2, pos=0.595,tick=F)
dev.off()
I have optimized the placement of the axis label and axis name using the line=1.5 and pos=0.595 parameters within the mtext and axis functions, respectively. However, once I generate a separate plot with a different width (i.e. width=8.0), the placement no longer resembles the physical distance that I specified in the original graph using a width of 5.0. So the questions is - how do I specify these parameters so that the label and name will be exactly the same distance from the graph boundary regardless of the plot width. Could this be specified using a universal unit (i.e. pixels)?
Thanks in advance!
try grconvertX but you have to keep track of how you need the output. mtext(..., line = ) is in lines whereas axis(..., pos = ) is expecting user coordinates. Also you would need to use grconvertY where appropriate (ie, on sides 1 and 3)
pdf('test1.pdf',height=3.5,width=5.0)
plot(1,ylab=NA,yaxt='n',ann=F)
mtext('Y-Axis Label',side=2,line=1.5)
axis(2, pos=0.595,tick=F)
line <- grconvertX(1.5, 'lines', 'device')
pos <- grconvertX(0.595, 'user', 'device')
dev.off()
pdf('test1-again.pdf',height=3.5,width=5)
plot(1,ylab=NA,yaxt='n',ann=F)
mtext('Y-Axis Label',side=2, line = grconvertX(line, 'device', 'lines'))
axis(2, pos = grconvertX(pos, 'device', 'user'), tick = FALSE)
dev.off()
pdf('test1-wide.pdf',height=3.5,width=8)
plot(1,ylab=NA,yaxt='n',ann=F)
mtext('Y-Axis Label',side=2, line = grconvertX(line, 'device', 'lines'))
axis(2, pos = grconvertX(pos, 'device', 'user'), tick = FALSE)
dev.off()
When I manually add the following labels with (axis(1, at=1:27, labels=labs[0:27])):
> labs[0:27]
[1] "0\n9.3%" "1\n7.6%" "2\n5.6%" "3\n5.1%" "4\n5.7%" "5\n6.5%" "6\n7.3%" "7\n7.6%" "8\n7.5%" "9\n7%" "10\n6.2%" "11\n5.2%"
[13] "12\n4.2%" ........
I get the following:
How do I force all labels to be drawn so 1,3,5,6, and 11 are not skipped? (also, for extra credit, how do I shift the whole thing down a few pixels?)
If you want to force all labels to display, even when they are very close or overlapping, you can "trick" R into displaying them by adding odd and even axis labels with separate calls to the axis function, as follows:
labs <-c("0\n9.3%","1\n7.6%","2\n5.6%","3\n5.1%","4\n5.7%","5\n6.5%","6\n7.3%",
"7\n7.6%","8\n7.5%","9\n7%", "10\n6.2%","11\n5.2%","12\n4.2%",13:27)
n=length(labs)
plot(1:28, xaxt = "n")
axis(side=1, at=seq(1,n,2), labels=labs[seq(1,n,2)], cex.axis=0.6)
axis(side=1, at=seq(2,n,2), labels=labs[seq(2,n,2)], cex.axis=0.6)
You can play with cex.axis to get the text size that you want. Note, also, that you may have to adjust the number of values in at= and/or labels= so that they are equal.
I agree with #PLapointe and #joran that it's generally better not to tamper with R's default behavior regarding overlap. However, I've had a few cases where axis labels looked fine even when they weren't quite a full "m-width" apart, and I hit on the trick of alternating odd and even labels as a way to get the behavior I wanted.
?axis tells you that:
The code tries hard not to draw overlapping tick labels, and so will omit labels where they would abut or overlap previously drawn labels. This can result in, for example, every other tick being labelled. (The ticks are drawn left to right or bottom to top, and space at least the size of an ‘m’ is left between labels.)
Play with cex.axis so that labels are small enough to fit without overlapping
labs <-c("0\n9.3%","1\n7.6%","2\n5.6%","3\n5.1%","4\n5.7%","5\n6.5%","6\n7.3%",
"7\n7.6%","8\n7.5%","9\n7%", "10\n6.2%","11\n5.2%","12\n4.2%",12:27)
plot(1:27,xaxt = "n")
axis(side=1, at=1:27, labels=labs[0:27],cex.axis=0.35)
If you widen you graph (manually by dragging or programmatically), you can increase the size of your labels.
Although there are some good answers here, the OP didn't want to resize the labels or change anything about the plot besides fitting all of the axis labels. It's annoying, since often there appears to be plenty of room to fit all of the axis labels.
Here's another solution. Draw the plot without the axis, then add ticks with empty labels. Store the positions of the ticks in an object, so then you can go through each one and place it in the correct position on the axis.
plot(1:10, 1:10, yaxt = "n")
axis_ticks = axis(2, axTicks(2), labels = rep("", length(axTicks(2))))
for(i in axis_ticks) axis(2, i)
#PLapointe just posted what I was going to say, but omitted the bonus answer.
Set padj = 0.5 in axis to move the labels down slightly.
Perhaps draw and label one tick at a time, by calling axis repeatedly using mapply...
For example, consider the following data:
x = runif(100)*20
y = 10^(runif(100)*3)
The formula for y might look a bit odd; it gives random numbers distributed across three orders of magnitude such that the data will be evenly distributed on a plot where the y axis is on a log scale. This will help demonstrate the utility of axTicks() by calculating nice tick locations for us on a logged axis.
By default:
plot(x, y, log = "y")
returns:
Notice that 100 and 1000 labels are missing.
We can instead use:
plot(x, y, log = "y", yaxt = "n")
mapply(axis, side = 2, at = axTicks(2), labels = axTicks(2))
which calls axis() once for each tick location returned by axTicks(), thus plotting one tick at a time. The result:
What I like about this solution is that is uses only one line of code for drawing the axis, it prints exactly the default axis R would have made, except all ticks are labeled, and the labels don't go anywhere when the plot is resized:
I can't say the axis is useful in the resized example, but it makes the point about axis labels being permanent!
For the first (default) plot, note that R will recalculate tick locations when resizing.
For the second (always labeled) plot, the number and location of tick marks are not recalculated when the image is resized. The axis ticks calculated by axTicks depend upon the size of the display window when the plot is first drawn.
If you want want to force specific tick locations, try something like:
plot(x, y, log = "y", yaxt = "n")
mapply(axis, side = 2, at = c(1,10,100, 1000), labels = c("one", "ten", "hundred", "thousand"))
which yields:
axis() includes a gap.axis parameter that controls when labels are omitted. Setting this to a very negative number will force all labels to display, even if they overlap.
The padj parameter of axis() controls the y offset whilst plotting an individual axis.
par(mgp = c(3, 2, 0) will adjust the position of all axis labels for the duration of a plotting session: the second value (here 2, default 1) controls the position of the labels.
# Set axis text position, including for Y axis
par(mgp = c(3, 2, 0))
# Plot
plot(1:12, 1:12, log = 'x', ann = FALSE, axes = FALSE)
# Some numbers not plotted:
axis(1, 1:12)
# All numbers plotted, with manual offset
axis(1, 1:12, gap.axis = -100, padj = 0.5)
I had a similar problem where I wanted to stagger the labels and get them to print without losing some. I created two sets of ticks showing second set below the other to make it look as if it staggers.
xaxis_stagger = function(positions,labels) {
odd=labels[seq(1,length(labels),2)]
odd_pos=positions[seq(1,length(positions),2)]
even=labels[seq(2,length(labels),2)]
even_pos=positions[seq(2,length(positions),2)]
axis(side=1,at=odd_pos,labels=odd)
axis(side=1,at=even_pos,labels=even,padj=1.5)
}
So you give the positions where you want the ticks to be and the labels for those ticks and this would then re-organise it into two sets of axis and plot them on the original plot. Original plot would be done with xaxt="n".
Here is shown how to label histogram bars with data values or percents using labels = TRUE. Is it also possible to rotate those labels? My goal is to rotate them to 90 degrees because now the labels over bars overrides each other and it is unreadable.
PS: please note that my goal is not to rotate y-axis labels as it is shown e.g. here
Using mtcars, here's one brute-force solution (though it isn't very brutish):
h <- hist(mtcars$mpg)
maxh <- max(h$counts)
strh <- strheight('W')
strw <- strwidth(max(h$counts))
hist(mtcars$mpg, ylim=c(0, maxh + strh + strw))
text(h$mids, strh + h$counts, labels=h$counts, adj=c(0, 0.5), srt=90)
The srt=90 is the key here, rotating 90 degrees counter-clockwise (anti-clockwise?).
maxh, strh, and strw are used (1) to determine how much to extend the y-axis so that the text is not clipped to the visible figure, and (2) to provide a small pad between the bar and the start of the rotated text. (The first reason could be mitigated by xpd=TRUE instead, but it might impinge on the main title, and will be a factor if you set the top margin to 0.)
Note: if using density instead of frequency, you should use h$density instead of h$counts.
Edit: changed adj, I always forget the x/y axes on it stay relative to the text regardless of rotation.
Edit #2: changing the first call to hist so the string height/width are calculate-able. Unfortunately, plotting twice is required in order to know the actual height/width.
I'm trying to figure out a way to calculate the height of a legend for a plot prior to setting the margins of the plot. I intend to place the legend below the plot below the x-axis labels and title.
As it is part of a function which plots a range of things the legend can grow and shrink in size to cater for 2 items, up to 15 or more, so I need to figure out how I can do this dynamically rather that hard-coding. So, in the end I need to dynamically set the margin and some other bits and pieces.
The key challenge is to figure out the height of the legend to feed into par(mar) prior to drawing the plot, but after dissecting the base codes for legend however, it seems impossible to get a solid estimate of the height value unless the plot is actually drawn (chicken and egg anyone?)
Here's what I've tried already:
get a height using the legend$rect$h output from the base legend function (which seems to give a height value which is incorrect unless the plot is actually drawn)
calculate the number of rows in the legend (easy) and multiply this by the line height (in order to do this, seems you'd need to translate into inches (the base legend code uses yinch and I've also tried grconvertY but neither of those work unless a plot has been drawn).
Another challenge is to work out the correct y value for placement of the legend - I figure that once I've solved the first challenge, the second will be easy.
EDIT:
After a day of sweating over how this is (not) working. I have a couple of insights and a couple of questions. For the sake of clarity, this is what my function essentially does:
step 1) set the margins
step 2) create the barplot on the left axis
step 3) re-set the usr coordinates - this is necessary to ensure alignment of the right axis otherwise it plots against the x-axis scale. Not good when they are markedly different.
step 4) create the right axis
step 5) create a series of line charts on the right axis
step 6) do some labelling of the two axes and the x-axis
step 7) add in the legend
Here are the questions
Q1) What units are things reported in? I'm interested in margin lines and coordinates (user-coordinates), inches is self explanatory. - I can do some conversions using grconvertY() but I'm not sure what I'm looking at and what I should be converting to - the documentation isn't so great.
Q2) I need to set the margin in step 1 so that there is enough room at the bottom of the chart for the legend. I think I'm getting that right, however I need to set the legend after the right axis and line charts are set, which means that the user coordinates (and the pixel value of an inch, has changed. Because of Q1 above I'm not sure how to translate one system to the other. Any ideas in this regard would be appreciated.
After another day of sweating over this here's what solved it mostly for me.
I pulled apart the code for the core legend function and compiled this:
#calculate legend buffer
cin <- par("cin")
Cex <- par("cex")
yc <- Cex * cin[2L] #cin(inches) * maginfication
yextra <- 0
ymax <- yc * max(1, strheight("Example", units = "inches", cex = Cex)/yc)
ychar <- yextra + ymax #coordinates
legendHeight <- (legendLines * ychar) + yc # in
Which is essentially mimicking the way the core function calculates legend height but returns the height in inches rather than in user coordinates. legendLines is the number of lines in the legend.
After that, it's a doddle to work out how to place the legend, and to set the lower margin correctly. I'm using:
#calculate inches per margin line
inchesPerMarLine<-par("mai")[1]/par("mar")[1]
To calculate the number of inches per margin line, and the following to set the buffers (for the axis labels and title, and the bottom of the chart), and the margin of the plot.
#set buffers
bottomBuffer = 1
buffer=2
#calculate legend buffer
legBuffer <- legendHeight/inchesPerMarLine
#start the new plot
plot.new()
# set margin
bottomMargin <- buffer + legBuffer + bottomBuffer
par(mar=c(bottomMargin,8,3,5))
The plot is made
barplot(data, width=1, col=barCol, names.arg=names, ylab="", las=1 ,axes=F, ylim=c(0,maxL), axis.lty=1)
And then the legend is placed. I've used a different method to extract the legend width which does have some challenges when there is a legend with 1 point, however, it works ok for now. Putting the legend into a variable allows you to access the width of the box like l$rect$w. trace=TRUE and plot=FALSE stop the legend being written to the plot just yet.
ycoord <- -1*(yinch(inchesPerMarLine*buffer)*1.8)
l<-legend(x=par("usr")[1], y=ycoord, inset=c(0,-0.25), legendText, fill=legendColour, horiz=FALSE, bty = "n", ncol=3, trace=TRUE,plot=FALSE)
lx <- mean(par("usr")[1:2]-(l$rect$w/2))
legend(x=lx, y=ycoord, legendText, fill=legendColour, horiz=FALSE, bty = "n", ncol=3)
For completeness, this is how I calculate the number of lines in the legend. Note - the number of columns in the legend is 3. labelSeries is the list of legend labels.
legendLines <- ceiling(nrow(labelSeries)/3)
I have a plot that has data that runs into the area I'd like to use for a legend. Is there a way to have the plot automatically put in something like a header space above the highest data points to fit the legend into?
I can get it to work if I manually enter the ylim() arguments to expand the size and then give the exact coordinates of where I want the legend located, but I'd prefer to have a more flexible means of doing this as it's a front end for a data base query and the data levels could have very different levels.
Edit 2017:
use ggplot and theme(legend.position = ""):
library(ggplot2)
library(reshape2)
set.seed(121)
a=sample(1:100,5)
b=sample(1:100,5)
c=sample(1:100,5)
df = data.frame(number = 1:5,a,b,c)
df_long <- melt(df,id.vars = "number")
ggplot(data=df_long,aes(x = number,y=value, colour=variable)) +geom_line() +
theme(legend.position="bottom")
Original answer 2012:
Put the legend on the bottom:
set.seed(121)
a=sample(1:100,5)
b=sample(1:100,5)
c=sample(1:100,5)
dev.off()
layout(rbind(1,2), heights=c(7,1)) # put legend on bottom 1/8th of the chart
plot(a,type='l',ylim=c(min(c(a,b,c)),max(c(a,b,c))))
lines(b,lty=2)
lines(c,lty=3,col='blue')
# setup for no margins on the legend
par(mar=c(0, 0, 0, 0))
# c(bottom, left, top, right)
plot.new()
legend('center','groups',c("A","B","C"), lty = c(1,2,3),
col=c('black','black','blue'),ncol=3,bty ="n")
You have to add the size of the legend box to the ylim range
#Plot an empty graph and legend to get the size of the legend
x <-1:10
y <-11:20
plot(x,y,type="n", xaxt="n", yaxt="n")
my.legend.size <-legend("topright",c("Series1","Series2","Series3"),plot = FALSE)
#custom ylim. Add the height of legend to upper bound of the range
my.range <- range(y)
my.range[2] <- 1.04*(my.range[2]+my.legend.size$rect$h)
#draw the plot with custom ylim
plot(x,y,ylim=my.range, type="l")
my.legend.size <-legend("topright",c("Series1","Series2","Series3"))
Building on #P-Lapointe solution, but making it extremely easy, you could use the maximum values from your data using max() and then you re-use those maximum values to set the legend xy coordinates. To make sure you don't get beyond the borders, you set up ylim slightly over the maximum values.
a=c(rnorm(1000))
b=c(rnorm(1000))
par(mfrow=c(1,2))
plot(a,ylim=c(0,max(a)+1))
legend(x=max(a)+0.5,legend="a",pch=1)
plot(a,b,ylim=c(0,max(b)+1),pch=2)
legend(x=max(b)-1.5,y=max(b)+1,legend="b",pch=2)
?legend will tell you:
Arguments
x, y
the x and y co-ordinates to be used to position the legend. They can be specified by keyword or in any way which is accepted by xy.coords: See ‘Details’.
Details:
Arguments x, y, legend are interpreted in a non-standard way to allow the coordinates to be specified via one or two arguments. If legend is missing and y is not numeric, it is assumed that the second argument is intended to be legend and that the first argument specifies the coordinates.
The coordinates can be specified in any way which is accepted by xy.coords. If this gives the coordinates of one point, it is used as the top-left coordinate of the rectangle containing the legend. If it gives the coordinates of two points, these specify opposite corners of the rectangle (either pair of corners, in any order).
The location may also be specified by setting x to a single keyword from the list bottomright, bottom, bottomleft, left, topleft, top, topright, right and center. This places the legend on the inside of the plot frame at the given location. Partial argument matching is used. The optional inset argument specifies how far the legend is inset from the plot margins. If a single value is given, it is used for both margins; if two values are given, the first is used for x- distance, the second for y-distance.