I am trying to combine two ggplot objects with patchwork
- two plots with different subsets of data, but the same x variable (and therefore same unit). I would like to align the plots according to the x values - Each x unit should have the same physical width in the final plot.
This is very easy when actually plotting the entire width of the larger data set (see plot below) - but I struggle to plot only parts of the data and keeping the same alignment.
library(ggplot2)
library(patchwork)
library(dplyr)
p1 <-
ggplot(mtcars, aes(mpg)) +
geom_density(trim = TRUE) +
scale_x_continuous(limits = c(10,35))
p2 <-
ggplot(filter(mtcars, mpg < 20), aes(mpg)) +
geom_histogram(binwidth = 1, boundary = 1) +
scale_x_continuous(limits = c(10,35))
p1/p2
Created on 2019-08-07 by the reprex package (v0.3.0)
The desired output
That's photoshopped
adding coord_cartesian(xlim = c(10,(20 or 35)), clip = 'off')
, and/or changing scale_x
limits to c(0,(20 or 35))
doesn't work.
patchwork
also won't let me set the widths of both plots when they are in two rows, which makes sense in a way. So I could create an empty plot for the second row and set the widths for those, but this seems a terrible hack and I feel there must be a much easier solution.
I am not restricted to patchwork
, but any solution allowing to use it would be very welcome.
I modified the align_plots
function from the cowplot package for this, so that its plot_grid
function can now support adjustments to the dimensions of each plot.
(The main reason I went with cowplot rather than patchwork is that I haven't had much tinkering experience with the latter, and overloading common operators like +
makes me slightly nervous.)
Demonstration of results
# x / y axis range of p1 / p2 have been changed for illustration purpose
p1 <- ggplot(mtcars, aes(mpg, 1 + stat(count))) +
geom_density(trim = TRUE) +
scale_x_continuous(limits = c(10,35)) +
coord_cartesian(ylim = c(1, 3.5))
p2 <- ggplot(filter(mtcars, mpg >= 15 & mpg < 30), aes(mpg)) +
geom_histogram(binwidth = 1, boundary = 1)
plot_grid(p1, p2, ncol = 1, align = "v") # plots in 1 column, x-axes aligned
plot_grid(p1, p2, nrow = 1, align = "h") # plots in 1 row, y-axes aligned
Plots in 1 column (x-axes aligned for 15-28 range):
Plots in 1 row (y-axes aligned for 1 - 3.5 range):
Caveats
This hack assumes the plots that the user intends to align (either horizontally or vertically) have reasonably similar axes of comparable magnitude. I haven't tested it on more extreme cases.
This hack expects simple non-faceted plots in Cartesian coordinates. I'm not sure what one could expect from aligning faceted plots. Similarly, I'm not considering polar coordinates (what's there to align?) or map projections (haven't looked into this, but they feel rather complicated).
This hack expects the gtable cell containing the plot panel to be in the 7th row / 5th column of the gtable object, which is based on my understanding of how ggplot objects are typically converted to gtables, and may not survive changes to the underlying code.
Code
Modified version of cowplot::align_plots
:
align_plots_modified <- function (..., plotlist = NULL, align = c("none", "h", "v", "hv"),
axis = c("none", "l", "r", "t", "b", "lr", "tb", "tblr"),
greedy = TRUE) {
plots <- c(list(...), plotlist)
num_plots <- length(plots)
grobs <- lapply(plots, function(x) {
if (!is.null(x)) as_gtable(x)
else NULL
})
halign <- switch(align[1], h = TRUE, vh = TRUE, hv = TRUE, FALSE)
valign <- switch(align[1], v = TRUE, vh = TRUE, hv = TRUE, FALSE)
vcomplex_align <- hcomplex_align <- FALSE
if (valign) {
# modification: get x-axis value range associated with each plot, create union of
# value ranges across all plots, & calculate the proportional width of each plot
# (with white space on either side) required in order for the plots to align
plot.x.range <- lapply(plots, function(x) ggplot_build(x)$layout$panel_params[[1]]$x.range)
full.range <- range(plot.x.range)
plot.x.range <- lapply(plot.x.range,
function(x) c(diff(c(full.range[1], x[1]))/ diff(full.range),
diff(x)/ diff(full.range),
diff(c(x[2], full.range[2]))/ diff(full.range)))
num_widths <- unique(lapply(grobs, function(x) {
length(x$widths)
}))
num_widths[num_widths == 0] <- NULL
if (length(num_widths) > 1 || length(grep("l|r", axis[1])) > 0) {
vcomplex_align = TRUE
warning("Method not implemented for faceted plots. Placing unaligned.")
valign <- FALSE
}
else {
max_widths <- list(do.call(grid::unit.pmax,
lapply(grobs, function(x) {x$widths})))
}
}
if (halign) {
# modification: get y-axis value range associated with each plot, create union of
# value ranges across all plots, & calculate the proportional width of each plot
# (with white space on either side) required in order for the plots to align
plot.y.range <- lapply(plots, function(x) ggplot_build(x)$layout$panel_params[[1]]$y.range)
full.range <- range(plot.y.range)
plot.y.range <- lapply(plot.y.range,
function(x) c(diff(c(full.range[1], x[1]))/ diff(full.range),
diff(x)/ diff(full.range),
diff(c(x[2], full.range[2]))/ diff(full.range)))
num_heights <- unique(lapply(grobs, function(x) {
length(x$heights)
}))
num_heights[num_heights == 0] <- NULL
if (length(num_heights) > 1 || length(grep("t|b", axis[1])) > 0) {
hcomplex_align = TRUE
warning("Method not implemented for faceted plots. Placing unaligned.")
halign <- FALSE
}
else {
max_heights <- list(do.call(grid::unit.pmax,
lapply(grobs, function(x) {x$heights})))
}
}
for (i in 1:num_plots) {
if (!is.null(grobs[[i]])) {
if (valign) {
grobs[[i]]$widths <- max_widths[[1]]
# modification: change panel cell's width to a proportion of unit(1, "null"),
# then add whitespace to the left / right of the plot's existing gtable
grobs[[i]]$widths[[5]] <- unit(plot.x.range[[i]][2], "null")
grobs[[i]] <- gtable::gtable_add_cols(grobs[[i]],
widths = unit(plot.x.range[[i]][1], "null"),
pos = 0)
grobs[[i]] <- gtable::gtable_add_cols(grobs[[i]],
widths = unit(plot.x.range[[i]][3], "null"),
pos = -1)
}
if (halign) {
grobs[[i]]$heights <- max_heights[[1]]
# modification: change panel cell's height to a proportion of unit(1, "null"),
# then add whitespace to the bottom / top of the plot's existing gtable
grobs[[i]]$heights[[7]] <- unit(plot.y.range[[i]][2], "null")
grobs[[i]] <- gtable::gtable_add_rows(grobs[[i]],
heights = unit(plot.y.range[[i]][1], "null"),
pos = -1)
grobs[[i]] <- gtable::gtable_add_rows(grobs[[i]],
heights = unit(plot.y.range[[i]][3], "null"),
pos = 0)
}
}
}
grobs
}
Utilising the above modified function with cowplot package's plot_grid
:
# To start using (in current R session only; effect will not carry over to subsequent session)
trace(cowplot::plot_grid, edit = TRUE)
# In the pop-up window, change `grobs <- align_plots(...)` (at around line 27) to
# `grobs <- align_plots_modified(...)`
# To stop using
untrace(cowplot::plot_grid)
(Alternatively, we can define a modified version of plot_grid
function that uses align_plots_modified
instead of cowplot::align_plots
. Results would be the same either way.)
Here is an option with grid.arrange
that does not use a blank plot, but requires a manual of adjustment of:
library(ggplot2)
library(dplyr)
library(gridExtra)
p1 <-
ggplot(mtcars, aes(mpg)) +
geom_density(trim = TRUE) +
scale_x_continuous(limits = c(10,35), breaks=seq(10,35,5), expand = expand_scale(add=c(0,0)))
p2 <-
ggplot(filter(mtcars, mpg < 20), aes(mpg)) +
geom_histogram(binwidth = 1, boundary = 1) +
scale_x_continuous(limits = c(10,20), breaks=seq(10,20,5), expand = expand_scale(add=c(0,0))) +
scale_y_continuous(labels = scales::number_format(accuracy = 0.01)) +
theme(plot.margin = unit(c(0,1,0,0), "cm"))
grid.arrange(p1, p2,
layout_matrix = rbind(c(1, 1), c(2, NA))
)
Should make this plot:
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With