Related
I've got a few different categories that I want to plot. These are different categories, each with their own set of labels, but which makes sense to group together in the document. The following gives some simple stacked bar chart examples:
df <- data.frame(x=c("a", "b", "c"),
y=c("happy", "sad", "ambivalent about life"))
ggplot(df, aes(x=factor(0), fill=x)) + geom_bar()
ggplot(df, aes(x=factor(0), fill=y)) + geom_bar()
The problem is that with different labels, the legends have different widths, which means the plots have different widths, leading to things looking a bit goofy if I make a table or \subfigure elements. How can I fix this?
Is there a way to explicitly set the width (absolute or relative) of either the plot or the legend?
Edit: Very easy with egg package
# install.packages("egg")
library(egg)
p1 <- ggplot(data.frame(x=c("a","b","c"),
y=c("happy","sad","ambivalent about life")),
aes(x=factor(0),fill=x)) +
geom_bar()
p2 <- ggplot(data.frame(x=c("a","b","c"),
y=c("happy","sad","ambivalent about life")),
aes(x=factor(0),fill=y)) +
geom_bar()
ggarrange(p1,p2, ncol = 1)
Original Udated to ggplot2 2.2.1
Here's a solution that uses functions from the gtable package, and focuses on the widths of the legend boxes. (A more general solution can be found here.)
library(ggplot2)
library(gtable)
library(grid)
library(gridExtra)
# Your plots
p1 <- ggplot(data.frame(x=c("a","b","c"),y=c("happy","sad","ambivalent about life")),aes(x=factor(0),fill=x)) + geom_bar()
p2 <- ggplot(data.frame(x=c("a","b","c"),y=c("happy","sad","ambivalent about life")),aes(x=factor(0),fill=y)) + geom_bar()
# Get the gtables
gA <- ggplotGrob(p1)
gB <- ggplotGrob(p2)
# Set the widths
gA$widths <- gB$widths
# Arrange the two charts.
# The legend boxes are centered
grid.newpage()
grid.arrange(gA, gB, nrow = 2)
If in addition, the legend boxes need to be left justified, and borrowing some code from here written by #Julius
p1 <- ggplot(data.frame(x=c("a","b","c"),y=c("happy","sad","ambivalent about life")),aes(x=factor(0),fill=x)) + geom_bar()
p2 <- ggplot(data.frame(x=c("a","b","c"),y=c("happy","sad","ambivalent about life")),aes(x=factor(0),fill=y)) + geom_bar()
# Get the widths
gA <- ggplotGrob(p1)
gB <- ggplotGrob(p2)
# The parts that differs in width
leg1 <- convertX(sum(with(gA$grobs[[15]], grobs[[1]]$widths)), "mm")
leg2 <- convertX(sum(with(gB$grobs[[15]], grobs[[1]]$widths)), "mm")
# Set the widths
gA$widths <- gB$widths
# Add an empty column of "abs(diff(widths)) mm" width on the right of
# legend box for gA (the smaller legend box)
gA$grobs[[15]] <- gtable_add_cols(gA$grobs[[15]], unit(abs(diff(c(leg1, leg2))), "mm"))
# Arrange the two charts
grid.newpage()
grid.arrange(gA, gB, nrow = 2)
Alternative solutions There are rbind and cbind functions in the gtable package for combining grobs into one grob. For the charts here, the widths should be set using size = "max", but the CRAN version of gtable throws an error.
One option: It should be obvious that the legend in the second plot is wider. Therefore, use the size = "last" option.
# Get the grobs
gA <- ggplotGrob(p1)
gB <- ggplotGrob(p2)
# Combine the plots
g = rbind(gA, gB, size = "last")
# Draw it
grid.newpage()
grid.draw(g)
Left-aligned legends:
# Get the grobs
gA <- ggplotGrob(p1)
gB <- ggplotGrob(p2)
# The parts that differs in width
leg1 <- convertX(sum(with(gA$grobs[[15]], grobs[[1]]$widths)), "mm")
leg2 <- convertX(sum(with(gB$grobs[[15]], grobs[[1]]$widths)), "mm")
# Add an empty column of "abs(diff(widths)) mm" width on the right of
# legend box for gA (the smaller legend box)
gA$grobs[[15]] <- gtable_add_cols(gA$grobs[[15]], unit(abs(diff(c(leg1, leg2))), "mm"))
# Combine the plots
g = rbind(gA, gB, size = "last")
# Draw it
grid.newpage()
grid.draw(g)
A second option is to use rbind from Baptiste's gridExtra package
# Get the grobs
gA <- ggplotGrob(p1)
gB <- ggplotGrob(p2)
# Combine the plots
g = gridExtra::rbind.gtable(gA, gB, size = "max")
# Draw it
grid.newpage()
grid.draw(g)
Left-aligned legends:
# Get the grobs
gA <- ggplotGrob(p1)
gB <- ggplotGrob(p2)
# The parts that differs in width
leg1 <- convertX(sum(with(gA$grobs[[15]], grobs[[1]]$widths)), "mm")
leg2 <- convertX(sum(with(gB$grobs[[15]], grobs[[1]]$widths)), "mm")
# Add an empty column of "abs(diff(widths)) mm" width on the right of
# legend box for gA (the smaller legend box)
gA$grobs[[15]] <- gtable_add_cols(gA$grobs[[15]], unit(abs(diff(c(leg1, leg2))), "mm"))
# Combine the plots
g = gridExtra::rbind.gtable(gA, gB, size = "max")
# Draw it
grid.newpage()
grid.draw(g)
The cowplot package also has the align_plots function for this purpose (output not shown),
both2 <- align_plots(p1, p2, align="hv", axis="tblr")
p1x <- ggdraw(both2[[1]])
p2x <- ggdraw(both2[[2]])
save_plot("cow1.png", p1x)
save_plot("cow2.png", p2x)
and also plot_grid which saves the plots to the same file.
library(cowplot)
both <- plot_grid(p1, p2, ncol=1, labels = c("A", "B"), align = "v")
save_plot("cow.png", both)
As #hadley suggests, rbind.gtable should be able to handle this,
grid.draw(rbind(ggplotGrob(p1), ggplotGrob(p2), size="last"))
however, the layout widths should ideally be size="max", which doesn't cope well with some types of grid units.
Just by chance, I noticed that Arun's solution he had suggested in his comments hasn't been picked up. I feel his simple and efficient approach is really worth to be illustrated.
Arun suggested to move the legend to the top or bottom:
ggplot(df, aes(x=factor(0), fill=x)) + geom_bar() + theme(legend.position = "bottom")
ggplot(df, aes(x=factor(0), fill=y)) + geom_bar() + theme(legend.position = "bottom")
Now, the plots have the same width as requested. In addition, the plot area is equally sized in both cases.
If there are more factors or even longer labels, it might become necessary to play around with the legend, e.g., to display the legend in two ore more rows. theme() and guide_legend() have several parameters to control the position and appearance of legends in ggplot2.
I created a little function based on the answer of #Sandy.
same.size.ggplot <- function(vector.string.graph, # a vector of strings which correspond to Robject ggplot graphs
reference.string.graph, # a string of a Robject ggplot graphs where height and/or height will be taken for reference
width = T, # if you wanna adapat only the width
height = F # if you wanna adapat only the height
) {
# example: same.size.ggplot(p0rep(c("a", "b"), thre), "a30")
which(vector.string.graph %in% reference.string.graph)
newref <- ggplotGrob(get(reference.string.graph))
ref.width <- newref$widths
ref.height <- newref$heights
assign(reference.string.graph, newref, env = parent.frame(1))
for(i in seq_along(vector.string.graph)) {
if(vector.string.graph[i] != reference.string.graph) {
new <- ggplotGrob(get(vector.string.graph[i]))
if( width ) {
new$widths <- ref.width
}
if( height ) {
new$heights <- ref.height
}
assign(vector.string.graph[i], new, env = parent.frame(1))
}
}
}
p1 <- ggplot(data.frame(x=c("a","b","c"),y=c("happy","sad","ambivalent about life")),aes(x=factor(0),fill=x)) + geom_bar()
p2 <- ggplot(data.frame(x=c("a","b","c"),y=c("happy","sad","ambivalent about life")),aes(x=factor(0),fill=y)) + geom_bar()
p3 <- ggplot(data.frame(x=c("a","b","c"),y=c("Crazy happy","sad","Just follow the flow")),aes(x=factor(0),fill=y)) + geom_bar()
grid.arrange(p1, p2, p3, ncol = 1)
same.size.ggplot(c("p1", "p2", "p3"), "p2") # same as same.size.ggplot(c("p2", "p3"), "p1")
grid.arrange(p1, p2, p3, ncol = 1)
Before
After
You could also use the patchwork-package for that:
require(ggplot2)
require(patchwork)
# data
df = data.frame(x = c("a", "b", "c"),
y = c("happy", "sad", "ambivalent about life"))
p1 = ggplot(df, aes(x=factor(0), fill=x)) + geom_bar()
p2 = ggplot(df, aes(x=factor(0), fill=y)) + geom_bar()
# Patchwork 1: Does it automatically
p1 / p2
# Patchwork 2: Create a list
l = patchwork::align_patches(p1, p2)
The code is as follows:
set.seed(123)
d1=data.frame(x=runif(10),y=runif(10),z=runif(10,1,10))
d2=data.frame(x=runif(10),y=runif(10),z=runif(10,100,1000))
ggplot()+geom_point(aes(x,y,size=z),data=d1)+
geom_line(aes(x,y,size=z),data=d2)
And the result is like this:
The size of points are too small so I want to change its size by scale_size. However, it seems both lines and points are influenced. So I wonder if there is a way to scale lines and points separately with a separate legend?
The two ways I can think of are 1) combining two legend grobs or 2) hacking another legend aesthetic. Both of these were mentioned by #Mike Wise in the comments above.
Approach #1: combining 2 separate legends in the same plot using grobs.
I used code from this answer to grab the legend. Baptiste's arrangeGrob vignette is a useful reference.
library(grid); library(gridExtra)
#Function to extract legend grob
g_legend <- function(a.gplot){
tmp <- ggplot_gtable(ggplot_build(a.gplot))
leg <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
legend <- tmp$grobs[[leg]]
legend
}
#Create plots
p1 <- ggplot()+ geom_point(aes(x,y,size=z),data=d1) + scale_size(name = "point")
p2 <- ggplot()+ geom_line(aes(x,y,size=z),data=d2) + scale_size(name = "line")
p3 <- ggplot()+ geom_line(aes(x,y,size=z),data=d2) +
geom_point(aes(x,y, size=z * 100),data=d1) # Combined plot
legend1 <- g_legend(p1)
legend2 <- g_legend(p2)
legend.width <- sum(legend2$width)
gplot <- grid.arrange(p3 +theme(legend.position = "none"), legend1, legend2,
ncol = 2, nrow = 2,
layout_matrix = rbind(c(1,2 ),
c(1,3 )),
widths = unit.c(unit(1, "npc") - legend.width, legend.width))
grid.draw(gplot)
Note for printing: use arrangeGrob() instead of grid.arrange(). I had to use png; grid.draw; dev.off to save the (arrangeGrob) plot.
Approach #2: hacking another aesthetic legend.
MilanoR has a great post on this, focusing on colour instead of size.
More SO examples: 1) discrete colour and 2) colour gradient.
#Create discrete levels for point sizes (because points will be mapped to fill)
d1$z.bin <- findInterval(d1$z, c(0,2,4,6,8,10), all.inside= TRUE) #Create bins
#Scale the points to the same size as the lines (points * 100).
#Map points to a dummy aesthetic (fill)
#Hack the fill properties.
ggplot()+ geom_line(aes(x,y,size=z),data=d2) +
geom_point(aes(x,y, size=z * 100, fill = as.character(z.bin)),data=d1) +
scale_size("line", range = c(1,5)) +
scale_fill_manual("points", values = rep(1, 10) ,
guide = guide_legend(override.aes =
list(colour = "black",
size = sort(unique(d1$z.bin)) )))
I'm a noob in programming, but you could try this methode. As you see, my code uses points and paths. I define a vector of the length of number of paths. My lines have the size 1. Then I add the sizes of my points at the back of that vector.
size_vec<-c(rep(1, length(unique(Data$Satellite))), 1.4, 4.6, 4.2, 5.5)
plot <- ggplot(data) +
geom_point(aes(x = x_cor, y = y_cor, shape=Type, size=Type)) +
geom_path(aes(x = x_cor, y = y_cor, group = Tour, size=factor(Satellite))) +
scale_size_manual(values = size_vec, guide ='none')
How could I add box around legend key only for color not for size while keeping both legend keys. Code below add boxes to both mappings.
x<-1:6;
y<-factor(2:7);
z<-1:6;
df <- data.frame(x,y,z)
ggplot(df, aes(x,y)) +
geom_point(aes(colour=y, size = z) ) +
theme(legend.key = element_rect(colour = '#bdbdbd', size = 0.6))
Here is one approach using the ggplot layout and gtable. It extracts the color legend from the layout, draws boxes around each key, re-assambles the legend, then inserts the legend back into the ggplot layout.
library(ggplot2)
library(gtable)
library(grid)
x<-1:6;
y<-factor(2:7);
z<-1:6;
df <- data.frame(x,y,z)
p = ggplot(df, aes(x,y)) +
geom_point(aes(colour=y, size = z) )
# get ggplot grob
gt = ggplotGrob(p)
# Get the combined legend
leg = gtable_filter(gt, "guide-box")
# The legend has two parts.
# Get the second part - the color legend
leg2 = leg$grobs[[1]]$grobs[[2]]
# Get the locations of the top of each box containing the legend keys
# in this legend's layout
rects <- leg2$layout$t[grepl("bg", leg2$layout$name)]
# Draw boxes around each key
for(i in rects) leg2 = gtable_add_grob(leg2, grid.rect(gp = gpar(col = '#bdbdbd', fill = NA)), t = i, l = 2)
# Insert new color legend back into the combined legend
leg$grobs[[1]]$grobs[2][[1]] <- leg2
# Insert combined legend back into ggplot grob
gt$grobs[gt$layout$name == "guide-box"][[1]] <- leg
# Draw it
grid.newpage()
grid.draw(gt)
Here is a second approach (based on #Baptiste's answer here) that draw two plots: one containing the size legend, and the other containing the color legend (with boxes around the keys). It then extracts the legends from each plot's layout, combines the two legends into a single legend, then inserts the combined legend back into one of the layouts.
library(ggplot2)
library(gtable)
library(grid)
x<-1:6;
y<-factor(2:7);
z<-1:6;
df <- data.frame(x,y,z)
p1 = ggplot(df, aes(x,y)) +
geom_point(aes(colour=y, size = z) ) +
scale_colour_discrete(guide = "none")
p2 = ggplot(df, aes(x,y)) +
geom_point(aes(colour=y, size = z) ) +
scale_size(guide = "none") +
theme(legend.key = element_rect(colour = '#bdbdbd', size = 0.6))
# Get ggplot grobs
gt1 = ggplotGrob(p1)
gt2 = ggplotGrob(p2)
# Get the legends
leg1 = gtable_filter(gt1, "guide-box")
leg2 = gtable_filter(gt2, "guide-box")
# Combine the legends
leg <- rbind(leg1[["grobs"]][[1]], leg2[["grobs"]][[1]], size = "first")
# Insert legend into g1 (or g2)
gt1$grobs[gt1$layout$name == "guide-box"][[1]] <- leg
# Draw it
grid.newpage()
grid.draw(gt1)
I was wondering if there is a way to have a common legend and at the same time identical plot widths for a multi-panel plot. I know how to get identical plot widths without a legend, or with a legend that is not vertically centered.
Here's Hadley's fix that I use for the common legend (https://github.com/hadley/ggplot2/wiki/Share-a-legend-between-two-ggplot2-graphs):
library(ggplot2)
library(gridExtra)
dsamp <- diamonds[sample(nrow(diamonds), 1000), ]
p1 <- qplot(carat, price, data=dsamp, colour=clarity)
p2 <- qplot(carat, price, data=dsamp, colour=clarity, geom="path")
g_legend<-function(p){
tmp <- ggplotGrob(p)
leg <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
legend <- tmp$grobs[[leg]]
return(legend)}
legend <- g_legend(p1)
lwidth <- sum(legend$width)
## using grid.arrange for convenience
## could also manually push viewports
grid.arrange(arrangeGrob(p1 + theme(legend.position="none"),
p2 + theme(legend.position="none"),
main ="this is a title",
left = "This is my global Y-axis title"), legend,
widths=unit.c(unit(1, "npc") - lwidth, lwidth), nrow=1)
To illustrate my point, I changed p2 by adding two other variables of the same dataset keeping clarity as the common legend:
p1 <- qplot(carat, price, data=dsamp, colour=clarity)
p2 <- qplot(depth, table, data=dsamp, colour=clarity, geom="path")
g_legend<-function(p){
tmp <- ggplotGrob(p)
leg <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
legend <- tmp$grobs[[leg]]
return(legend)}
legend <- g_legend(p1)
lwidth <- sum(legend$width)
## using grid.arrange for convenience
## could also manually push viewports
grid.arrange(arrangeGrob(p1 + theme(legend.position="none"),
p2 + theme(legend.position="none"),
main ="this is a title",
left = "This is my global Y-axis title"), legend,
widths=unit.c(unit(1, "npc") - lwidth, lwidth), nrow=1)
This may not make sense in this example, but I have four different measurements on the same experimental units, which I want to graphically display in 2 plots with same plot widths and a common vertically centered legend.
I think you can achieve your goal better with faceting:
ggplot(dsamp, aes(x = carat, y = price, colour = clarity)) +
geom_point(data = cbind(dsamp, group = 1)) +
geom_path(data = cbind(dsamp, group = 2)) +
facet_grid(group ~ .) +
theme(strip.text.y = element_blank(),
strip.background = element_blank()) +
ggtitle("this is a title")
I've assumed that you have identical variables in your plots as in your example. Otherwise, I've had to use a lot of tinkering when I needed to align plots myself: See scatterplot with alpha transparent histograms in R for an example.
Suppose I have two plots, side by side, with the same y-axis, generated by the following R code:
df <- data.frame(x=c(5,2,7,3), y=c(11,3,5,6), facet=c(1,1,2,2))
ggplot(df, aes(x, y)) + facet_grid(~facet) + geom_point()
Is it possible to write the y axis text (e.g., 10.0, 7.5, 5.0) in the middle, between the two plots? (Preferentially the text should be centered.)
Here is a way (well almost) using Baptiste's answer from this SO post Display y-axis for each subplot when faceting. Not quite in the middle but its close
library(ggplot2)
library(gtable)
# your data
df <- data.frame(x=c(5,2,7,3), y=c(11,3,5,6), facet=c(1,1,2,2))
# First plot (a bit of extra space between facets)
p <- ggplot(df, aes(x, y)) + facet_grid(~facet) +
geom_point() +
theme(panel.margin = unit(1, "lines"),
axis.text.y = element_text( hjust=0))
# get y-axis labels
g <- ggplotGrob(p)
axis <- gtable_filter(g, "axis-l")[["grobs"]][[1]][["children"]][["axis"]][,1]
# remove axis
g[["grobs"]][[4]][["children"]][["axis"]] <- NULL
# build plot & add axis to LHS of left facet
panels <- subset(g$layout, name == "panel")
g <- gtable_add_grob(g, grobs=axis, t = unique(panels$t), l=tail(panels$l, -1)-1)
grid.newpage()
grid.draw(g)