Advent of code 2. reusing incomplete layers.

[EvaMaeRey]

I think another advent question might about reusing layers that aren't fully locally defined. So you can grab layers from p1 and p2 below, which have layer information all local, and apply those layers in a new plot... (kind of inspired by the composition of swim plots in @rsh52's work - there are three layers and you might want to look at those in separate plots, and then superimpose them later)

library(ggplot2)
p1 <- ggplot() + 
  geom_point(data = mtcars,
             aes(mpg, mpg))

p2 <- ggplot() + 
  geom_point(aes(cty, hwy),
             data = mpg,
             color = "blue")


ggplot() + 
  p1$layers + 
  p2$layers + 
  coord_equal()

But using a more common plot building approach, we cannot just use the layer directly.

p1.2 <- ggplot(data = mtcars) + 
  geom_point(aes(mpg, mpg, color = NULL))

p1.3 <- ggplot(data = mtcars) + aes(cty, hwy), + 
  geom_point()


ggplot() + 
  p1.2$layers + 
  p2$layers + 
  coord_equal()
#> Don't know how to automatically pick scale for object of type
#> <tbl_df/tbl/data.frame>. Defaulting to continuous.
#> Don't know how to automatically pick scale for object of type
#> <tbl_df/tbl/data.frame>. Defaulting to continuous.
#> Error in `data[[var]][scale_index[[i]]]`:
#> ! Can't subset columns past the end.
#> ℹ Locations 12, 13, 14, …, 233, and 234 don't exist.
#> ℹ There are only 11 columns.

I think target might be add_layers() (via ggplot_add, but I haven't actually attempted).

ggplot()  + 
  add_layers(p1) + 
  add_layers(p2)  

[teunbrand]

Just commenting on the error message:

#> Don't know how to automatically pick scale for object of type
#> <tbl_df/tbl/data.frame>. Defaulting to continuous.

This is because you have aes(mpg), which will find mtcars$mpg when data = mtcars, but not when data = NULL (default). p1.2 and p1.3 only have global data, not local data, so aes(mpg) finds the mpg dataset rather than the mtcars$mpg.

[EvaMaeRey]

That makes sense. I think grabbing the plot data and adding it to the layer if not present, and pulling in mapping info if not locally specified. And returning that modified layer.

[EvaMaeRey]

So updating local data w/ global seems to work (using the p3 layer), but doing the same with mapping didn't (p4 layer). 🤔

library(ggplot2)
p3 <- ggplot(data = mtcars) + 
  geom_point(aes(mpg, mpg))

ggplot() + 
  p3$layers
#> Don't know how to automatically pick scale for object of type
#> <tbl_df/tbl/data.frame>. Defaulting to continuous.
#> Don't know how to automatically pick scale for object of type
#> <tbl_df/tbl/data.frame>. Defaulting to continuous.
#> Error in `data[[var]][scale_index[[i]]]`:
#> ! Can't subset columns past the end.
#> ℹ Locations 12, 13, 14, …, 233, and 234 don't exist.
#> ℹ There are only 11 columns.

p3$layers$geom_point$data <- p3$data

ggplot() + 
  p3$layers

p4 <- ggplot(data = mtcars) + 
  aes(cty, hwy) +
  geom_point()

ggplot() +
  p4$layers

p4$layers$geom_point$data <- p4$data
p4$layers$geom_point$mapping <- p4$mapping

ggplot() +
  p4$layers
#> Error in `geom_point()`:
#> ! Problem while computing aesthetics.
#> ℹ Error occurred in the 1st layer.
#> Caused by error:
#> ! object 'cty' not found

^{Created on 2025-01-07 with reprex v2.1.0}

[mjskay]

I think this is a really interesting question. Something that's bothered me with ggplot2 for quite awhile is that it doesn't have a formal notion of a vector of layers (which kind of relates to #66), which would be the obvious solution here in my opinion.

The lack of a "vector of layers" means that you get surprising results when you do things like add two layers together without a plot. E.g. take the following:

plot = ggplot()

layer1 = geom_point(data = mtcars, aes(mpg, mpg))
layer2 = geom_point(aes(cty, hwy), data = mpg, color = "blue")

The combined plot is:

plot + layer1 + layer2

We expect + to be associative, so we expect (plot + layer1) + layer2 to be equivalent to plot + (layer1 + layer2), but it isn't:

plot + (layer1 + layer2)
## Error in `+.gg`:
## ! Cannot add <ggproto> objects together.
## ℹ Did you forget to add this object to a <ggplot> object?
## Run `rlang::last_trace()` to see where the error occurred.

Because + is not defined for two layers, only a plot and a layer.

The existing solution is to explicitly create a list of layers, which acts sort of like a layer vector, a la plot + list(layer1, layer2). But I feel like it would be nicer to have an actual "layer vector". Then the solution to re-using layers would not be to try to extract a list of layers from an existing plot (or to create a list of layers), but to create a layer vector which could be used in several different plots.

[EvaMaeRey]

Collections of layers do feel a bit hard to manage. I guess I've though about this a little bit more from the layer-writing side. modify_layer() is cool. I think coming from a similar 'place' as the layers deletion idea - modify_layer seem a less destructive approach! 🙃

[teunbrand]

In dev ggplot2, layers will have (autogenerated) names

[EvaMaeRey]

like... 'layer-1'?

[EvaMaeRey]

just curious... layer-point-identity-identity-1 came to mind, but I guess you might want to be able to geom/stat/position swap, and then layer id changes?

[teunbrand]

They're just based on what you give as layer(name) and falls back to the constructor you used. You can't really decode the name back to reliable information.

[EvaMaeRey]

Somehow, this got me thinking about using patchwork with ggplot_add data_filter() #74 and then later aes() to get some in-paradigm 'crosstalk'-ish behavior, and learned about the * and & operators for applying new adds to patchwork plot ensembles.

Very exciting! 🤯 Though & it doesn't quite seem to be working w/ data_filter as written. I don't know if folks would have insights into that.

library(ggplot2)
data_filter <- function(.keep, .by) {
  structure(list(keep_specification = rlang::enquo(.keep), 
                 by_specification = rlang::enquo(.by)), 
            class = "filterobs")
}

ggplot_add.filterobs <- function(object, plot, object_name) {
  
  new_data <- dplyr::filter(plot$data, 
                            !!object$keep_specification, 
                            .by = !!object$by_specification)
  plot$data <- new_data
  plot

}

mtcars
plot(mtcars)

ggplot(mtcars) + 
  aes(factor(cyl), fill = factor(cyl)) + 
  geom_bar() -> p1;p1

ggplot(mtcars) + 
  aes(id = cyl, fill = factor(cyl)) + 
  ggcirclepack::geom_circlepack() + 
  coord_equal() -> p2

ggplot(mtcars) + 
  aes(wt, mpg, color = factor(cyl)) + 
  geom_point(size = 4) -> p3
  
library(patchwork)
p1 + p2 + p3

(p1 + p2 + p3) * data_filter(cyl != 4)

(p1 + p2 + p3) * aes(alpha = cyl != 4) 

The & operator is demonstrated w/ theme_minimal in the documentation...

patchwork <- p3 / (p1 | p2)
patchwork & theme_minimal()

And seems to work as advertised w/ aes, but not data_filter()

patchwork & aes(alpha = cyl == 4) & guides(alpha = "none")

patchwork & data_filter(cyl != 4)

Error in UseMethod("filter"): no applicable method for 'filter' applied to an object of class "waiver"

[EvaMaeRey]

https://r-graph-gallery.com/414-map-multiple-charts-in-ggiraph.html example of ggiraph interactive cross talk. I kind of wonder about ggiraph's potential usability w/ non-native layers like (i.e. extension layers like circlepack).

[teunbrand]

Have you ever approached ggiraph's author to give a talk? I'm also wondering how difficult it would be to build ggiraph-compatible layer extensions.

[EvaMaeRey]

He was on the call yesterday - planning on a June presentation tentatively. 🤩 @davidgohel