Basic one-sided formula support

[grantmcdermott]

quick examples:

devtools::load_all("~/Documents/Projects/plot2")
#> ℹ Loading plot2

par(pch = 16)

plot2(~ Temp, airquality)

plot2(~ Temp | Month, airquality, type = "b", par_restore = TRUE)

plot2(~ Temp | Day * Month, airquality, legend = FALSE)

^{Created on 2023-08-15 with reprex v2.0.2}

[grantmcdermott]

@zeileis do you mind taking a look at this? Note that this PR just provides basic one-side formula support and doesn't yet address #61. (I have some code ready to roll on top of this PR that does address #61, but it starts to get a little more complicated, so I want to hold back on it until we're happy with the basic scaffolding.)

[zeileis]

Nice! The only problem I saw was that the bylab title is wrong in the case where the are multiply by variables. In your example you had to switch off the legend because it got too long. A more manageable example could be:

airq <- transform(airquality, Temp = factor(Temp > 85, levels = c(FALSE, TRUE), labels = c("low", "high")))
plot2(~ Ozone | Temp + Month, data = airq, col = rep(palette.colors(5, "Tableau"), each = 2), pch = rep(c(1, 19), 5))

The fix would have just been to replace -(1L:2L) with -yxloc when putting together the bylab. However, when I went through the code I thought that I would add some more comments, add an explicit logical variable no_y, simplify the error message, and a couple of other small changes. They don't change the behavior of the code but I think it is a bit more readable now. Should I push this to the branch so that you can have a look? Or should I just make the minimal change?

[zeileis]

OK, I have pushed all my changes (including the cosmetic ones) now, I hope that's alright. Otherwise feel free to revert.

[grantmcdermott]

Excellent, thanks!

Btw, on the subject of legends that are too long, I have some ideas for continuous variable legends with interpolated (gradient) palette. I’ll create an issue and tag you when I get a sec to write up my thoughts.

[zeileis]

This would probably mean that we shouldn't just coerce all by variables to factors. The continuous gradient you mention could then be used for numeric by variables. However, the same idea/trick could not be used for other properties (pch or lty) that are necessarily discrete. Additionally/alternatively, one could be smarter in coercing numeric by variables to factors. Rather than just always using all unique variables via factor() one could cut() the variable at some quantiles - similar to what lattice does. My feeling was that this could easily get confusing, though, hence I did not try to do anything like that in the formula method.