Improvements to qualitative color scaling

[cpsievert]

scale_[fill/color]_discrete currently default to scale_[fill/color]_hue. While this is a great default when there are a large number of levels to encode, it's not a great default there are just a few (i.e., it's not colour-blind safe or aesthetically pleasing compared to carefully chosen qualitative palettes like colorbrewer or Okabe-Ito). In the event there isn't already a "blessed" ggplot2 scale (e.g., scale_colour_brewer()), the usual way to override the default is with a custom qualitative colorscale is to use scale_[fill/color]_manual(), but this also isn't ideal for situations where you don't want to make any assumptions about the number of levels to encode (i.e., scale_[fill/color]_manual() errors if there are more levels than color codes).

I'd like to propose that scale_[fill/color]_discrete gain a new argument, say codes, that would accept a vector of color codes. In the event codes is longer than the number of levels to encode, it would effectively be scale_[fill/color]_manual; otherwise, it would effectively be scale_[fill/color]_hue. Moreover, I think it would make sense for codes to default to something sensible like Okabe-Ito.

It seems the implementation for this would be relatively straight-forward (see here for a proof-of-concept). If there's interest in pursuing this idea, I'd be happy to start a PR.

[hadley]

Hmmm, I think I'd prefer to take the same approach as scale_colour_gradient(), i.e. making the default type a parameter that you could set. And then we'd also provide a colour scale that was colour blind friendly by default.

[cpsievert]

i.e. making the default type a parameter that you could set.

Yea, so something like codes = getOption("ggplot2.discrete.colour", default = "Okabe-Ito")?

[hadley]

I was thinking of a scale function

[clauswilke]

As discussed here, if we make this part of ggplot2 3.3.1 we shouldn't change the default color scales. Also, I'm not sure if @hadley's comment #3826 (comment) has been addressed in the PR.