Pagination of large facet plots

[MichaelHatherly]

When faceting a plot over a large number of columns/rows (or just wrapping with layout) the size of the plots can sometimes become quite unwieldy and requires manual intervention to increase the figure size or shrink the font size. Something that's a nice feature to have is pagination of faceted plots to reduce the amount of of axes that appear per-figure. There's a couple of options to do this with ggplot2: ggforce and gridExtra can handle this.

The basic look of the API is instead of writing this:

layer = data(df) * mapping(:id, :wt; layout = :dose => nonnumeric)
draw(layer)

we wrap the Layer with paginate:

draw(
	paginate(
		layer;
		limit = 4
	)
)

which just partitions the .data contained in the Layer based on what's used in layout (or row and col).

If there's interest in upstreaming this I'm happy to sort that out. Pagination has come in handy for a lot of our plotting needs in Pumas so I figured it may be more generally useful.

[piever]

Thanks, this looks very useful, it definitely would be great to upstream the feature to AoG.

About the API, draw now has a facet keyword argument to pass keywords specific to facet plots (see eg here), so maybe this could also be set there as a facet option.

OTOH, the current facet options tends to be things that are done directly via MakieLayout (the keywords are simply forwarded to facet!, which runs some Makie functions to improve the looks of a facet plot), so maybe it makes more sense that it is done wrapping the layer object.

[MichaelHatherly]

About the API, draw now has a facet keyword argument to pass keywords specific to facet plots (see eg here), so maybe this could also be set there as a facet option.

I considered that approach, the main issue I ran into with that approach, i.e.

draw(layers; facet = (; paginate = true,)) # or something like that..

is that the return type changes from FigureGrid to Vector{FigureGrid} by changing a keyword, which doesn't feel very nice. By operating on the Layer instead you can change the input argument type to something else (a Paginate object in the current iteration) which then dispatches to a different draw implementation which felt a bit better from an API point-of-view. It still feels a bit too close to some kind of "implicit broadcasting" behaviour and might be interesting to see whether a broadcasted draw might work

draw.(layers; kws...)

or just wrap it all in a separate paginate function instead of draw where

draw(
	paginate(
		layers;
		limit = 4
	)
)

would just become

paginate(layers; kws...)

though that does end up forcing the keywords associated with paginate into the same call as those for draw.

(Edit: the added benefit of just processing the Layer input is that the implementation of this doesn't end of needing to change any of the current internals of AoG, which was a nice added bonus.)

[piever]

Ah, of course, I hadn't thought that the output type is different. Then it makes a lot of sense to wrap the input, I agree that the API proposed initially, ie draw(paginate(layers; limit); kwargs...), is the cleanest solution.

[jkrumbiegel]

pagination has existed for a while now