Map multiple variables to different graphical parameters

[adrianadermon]

I would like to map, e.g., one grouping variable to color, and another to shape (in a scatter plot) or line type (in a line plot). This can be done with base plot, but is quite cumbersome. Is there a simple way to achieve it with tinyplot?

[zeileis]

Adrian, this is a good point and I don't think we have a fully automatic way for this, yet. But it's a bit less cumbersome than in base plots. You can easily add grouping variables in the formula y ~ x | z1 + z2 but you have to recycle your pch, lty, col arguments yourself, e.g., using rep().

library("tinyplot")
pal <- c("darkorange", "purple", "cyan4")
sym <- c(1, 16)
tinyplot(
  bill_dep ~ bill_len | sex + species,
  data = penguins,
  pch = rep(sym, 3),
  col = rep(pal, each = 2)
)

But maybe Grant @grantmcdermott has some more tricks for this, especially for getting two separate legends rather than only the legend for the interaction.

[grantmcdermott]

Dual legend support will be possible once #433 is merged... but only for the cex component.

I was actually thinking about this exact issue whilst preparing my userR! slides, since the ?penguins Examples in R 4.5.0 have the basic plot that Achim shows above.

One option could be a wholesale refactoring of the parameter logic (i.e., for col, pch, lty, lwd, and cex) so that these all accept NSE. In principle, this would let users conveniently specify, say, plt(y ~ x, data, col = <colvar>, pch = <pchvar>) and you'd get the appropriate mappings and legends (plural). BUT this would require a major refactoring of our internals and right now I have no idea of the tradeoffs versus our simple formula-based API. It's something I'd probably only attempt closer to tinyplot v1.0.0 release.

[grantmcdermott]

Related: #41 (with an ugly proof of concept implementation here: #343)

[zeileis]

You can probably guess, what my reaction is when it comes to extending NSE. I'd rather not go there, it's not fun.

Instead, this is what I would find easier to implement and to explain:

tinyplot(bill_dep ~ bill_len | sex + species, data = penguins, pch = "sex", col = "species")

Instead of already setting up interaction(sex, species) inside of the formula method, this would rather keep a two-column data.frame(sex, species) and pass this to the by argument of the default method.