geom_sf() and facet_*(..., scales="free"): plot aspect ratio

[adrfantini]

Setting scales="free" in facet_* resets geom_sf aspect ratio handling, even if not necessary. See reprex below.

library(ggplot2)
library(dplyr)
library(sf)
nc <- sf::st_read(system.file("shape/nc.shp", package = "sf"), quiet = TRUE) %>% mutate(dummy=1)
nc2 <- sf::st_read(system.file("shape/nc.shp", package = "sf"), quiet = TRUE) %>% mutate(dummy=2)
#Correct form factor for the plot:
ggplot(rbind(nc, nc2)) +  geom_sf(aes(fill = AREA)) + facet_wrap(~dummy)
#Adding scales="free" causes the plot to be warped:
ggplot(rbind(nc, nc2)) +  geom_sf(aes(fill = AREA)) + facet_wrap(~dummy, scales="free")

Without scales="free":

With scales="free":

scales="free" should allow facets to have different ranges without altering the plot aspect ratio.

[clauswilke]

I think the real bug is that coord_sf() now allows free scales. It's not generally possible to scale axes freely and maintain the correct aspect ratio and keep facet sizes equal.

I believe the bug was introduced as a result of #2594:

coord_sf()$is_free()
# [1] TRUE

versus:

coord_fixed()$is_free()
# [1] FALSE

The reason is coord_sf() inherits from coord_cartesian() but doesn't change the is_free() method:

ggplot2/R/sf.R

Line 262 in eecc450

CoordSf <- ggproto("CoordSf", CoordCartesian,

[adrfantini]

It's not generally possible to scale axes freely and maintain the correct aspect ratio and keep facet sizes equal.

What if I want to show some data over USA and Canada, in two different facets? Currently, (without scales = "free") I'll have two big facets spanning the whole of North America (One with Canada data, one with USA data), instead (using scales = "free") of two smaller facets spanning USA and Canada separately. Why shouldn't this be allowed?

[clauswilke]

The problem is that throughout the ggplot2 code base, the assumption is made that axes can be trained independently of other axes. To implement what you're requesting, one would have to write a training algorithm that trains x and y axes simultaneously while producing a targeted aspect ratio. This is possible in principle, but it's not immediately obvious how to do it in the current code base, at least to me. The best way forward might be to write a new faceting function specifically for this purpose, probably first in the form of an extension package. @thomasp85 may have further insight.

[batpigandme]

There are ggplot2 extensions (see http://www.ggplot2-exts.org/gallery/) that should allow you to arrange two maps side by side as such, e.g. patchwork, and (@clauswilke, correct me if I'm wrong) cowplot.

[clauswilke]

@batpigandme Yes, that's the other possible approach, make the plots separately and then arrange side-by-side. But then we don't get the power of faceting.

[batpigandme]

Oh yeah, of course, I was just offering a workaround!

[hadley]

Yes, this is a bug - this plot should throw an error message.

[adrfantini]

Thanks for the bugfix. I do not know the internals of ggplot so my request for clarification might seem trivial (and maybe already answered by @clauswilke ), but if the following works, why can't the same be made working with coord_sf?

d = data.frame(x=1:2)
ggplot(d, aes(x=x, y=x)) + geom_point() + facet_wrap(~x, scales="free")

[clauswilke]

coord_sf has a fixed aspect ratio. The appropriate comparison would be something like this:

library(ggplot2)
d = data.frame(x=c(1,100), y=c(1,100), f=c(1, 100))
ggplot(d, aes(x=y, y=y)) + geom_point() + facet_wrap(~f, scales="free") + coord_fixed()
#> Error: coord_fixed doesn't support free scales

Created on 2018-05-23 by the reprex package (v0.2.0).