Scale types overhaul

[hadley]

• Make sure all scales work with dates/datetimes
• Better colour/fill scales for ordered factors
• Warn when mapping factor to size
• Consider different viridis scales (using viridisLite) for fill/colour
• Better default breaks for log axes
• Document how to extend with your own S3 classes
• Consider support for percentage, proportion, and likert s3 types
• Could we have a jitter_na() type for position scales?

[huftis]

@hadley Regarding your last point, ‘different colour scales for fill/colour’: If you’re thinking about the continuous colour scales, I recommend watching this (fascinating!) video about the new default colour scale in matplotlib. (There’s also an R implementation.)

[hadley]

@huftis yup I'm aware of viridis, and I'm considering making it the default.

[hadley]

One hacky way of better display for missing values:

add_na <- function(x, prop = 0.1) {
  x[sample(seq_along(x) <= prop * length(x))] <- NA
  x
}

df <- tibble(
  x = letters[1:3] %>% sample(100, replace = TRUE) %>% add_na(),
  y = rnorm(100) %>% add_na()
)

ggplot(df, aes(x, y)) +
  geom_point()

miss <- df %>% filter(is.na(y))

# Jittering and display at one end of the scale is an old idea
# Andreas Buja and was implemented in XGobi/GGobi

ggplot(df, aes(x, y)) +
  geom_point(na.rm = TRUE) +
  geom_jitter(aes(y = 3.5), data = miss, width = 0.05, height = 0.05, colour = "darkred") +
  geom_hline(yintercept = 2.8, size = 10, colour = "white") +
  scale_y_continuous(
    breaks = c(-2, -1, 0, 1, 2, 3.5),
    labels = c(-2, -1, 0, 1, 2, "NA")
  )

[hadley]

Mostly done by @karawoo