Improve scales behavior for ordered factors

[karawoo]

Related to 1526. Ordered factors and regular factors currently behave pretty much the same, but this is not always appropriate. I've added new ordinal scales for size, alpha, and shape so that

• scale_shape_ordinal throws a warning (but scale_shape_discrete still does not)
• scale_size_ordinal does not throw a warning (but scale_size_discrete still does)
• scale_alpha_ordinal still does not throw a warning (but scale_alpha_discrete does)

Still to do:

• new colour/fill scales for ordinal variables (possibly with viridis)
• test the new behavior

[karawoo]

Some of this could be DRY-er, e.g.

scale_size_ordinal <- function(...) {
  suppressWarnings(scale_size_discrete(...))
}

might be better, or change the discrete version to

scale_size_discrete <- function(...) {
  warning("Using size for a discrete variable is not advised.", call. = FALSE)
  scale_size_ordinal(...)
}

[karawoo]

If we're going to go with viridis for ordinal scales then we may not need any additional functions, but if not then perhaps something like this should go in the scales package:

seq_ordinal_pal <- function(low, high) {
  function(n) {
    v <- seq(0, 1, length.out = n)
    scales::colour_ramp(c(low, high))(v)  
  }
}

Then scale_colour_ordinal would be:

scale_colour_ordinal <- function(..., low = "#132B43", high = "#56B1F7",
                                 na.value = "grey50") {
  discrete_scale(
    "colour",
    "sequence",
    seq_ordinal_pal(low, high),
    na.value = na.value,
    ...
    )
}

which produces output like this:

d <- diamonds[sample(nrow(diamonds), 1000), ]

ggplot(d, aes(carat, price)) +
  geom_point(aes(colour = cut))

[has2k1]

You need to be careful (or document the aberrations), depending on how you interpolate between the high and low colours of a colourmap, not all generated colours may be part of the continuous version of the colourmap.

This will definitely affect colourmaps that aim to perceptually-uniform, like viridis.

[karawoo]

@hadley GitHub isn't letting me request a review but can you tell me what you think of this? I can't remember if we decided on Monday what to do about the ordinal color scales.

[hadley]

Looks good so far. We should make a decision about ordinal colour scales too. Using viridis seems like a reasonable choice to me.

[karawoo]

Trying @jennybc's edits from #2189 in this PR to see if it makes codecov chill out.

Edit: wait is it now saying that my last commit (before merging the new changes) passed all the checks? I swear codecov was failing before...

[karawoo]

Using viridis sounds good to me, so once #2178 gets merged I can add that here.

[karawoo]

I removed Jenny's codecov-related commits and added scale_fill_ordinal and scale_colour_ordinal, which both default to viridis.

[karawoo]

@hadley one quick question before I merge -- since we're making viridis the default for ordered factors, should viridisLite go in Imports rather than Suggests?

[hadley]

Yes!

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/