Is there a reason that `gratia:::plot_penalty` uses `geom_raster` not `geom_tile`

[millerdl-CEH]

Hi Gavin,

I'm plotting some simple penalties with gratia::draw and they come out pretty icky because of geom_raster (i.e., what looks like anti-aliasing per cell but that blurs the whole matrix, at least with PDF output). I guess geom_tile will plot the cells crisply, but is slow for larger penalty matrices.

Would it be possible to supply a geom= option or somesuch for draw to allow the user to specify this? Also perhaps it's worth doing something where for matrices below a certain dimension it uses tile, but raster beyond that?

I can submit a patch if you think it's worth doing, I just wanted to check there wasn't some other reason before I do so.

cheers,
--dave

PS. yes, I am avoiding writing a talk.

[gavinsimpson]

It's the efficiency reason. You're seeing an antialiasing rendering issue in your pdf viewer. I see the same in some viewer on my Mac machines.

Do you know that geom_tile() avoids this behaviour? If so, I can look at options. Do the geoms use the same arguments/definitions? I recall that at least one of these geoms interprets things differently.

On the performance side, Eric was looking at this re your plot for NCV neighborhoods and geom_raster() was slower than geom_tile() IIRC in his tests. So I could be persuaded to switch.