The geom_hex geometry in ggplot2 colors hexagonal bins according to the number of points falling within them. This works pretty well for uniformly distributed data, but not so well if some regions are way more dense than others-- differences can get drowned out by the presence of a single very dense hexagon.
How can I make the density color scale use a log scale or some other kind of normalizing transformation?
ggplot 3.0+ demystifies the calculation of summary metrics via the new stat() internal function. This makes it easier to modify the statistic being used to create the fill for the hexes. So for example:
df <- data.frame(
x = rnorm(1000),
y = rnorm(1000)
)
plot.df <- ggplot(data = df, aes(x = x, y = y)) +
geom_hex(aes(fill = stat(count)))
print(plot.df)
plot.df.log <- ggplot(data = df, aes(x = x, y = y)) +
geom_hex(aes(fill = stat(log(count))))
print(plot.df.log)
In place of log, you could do any arbitrary transformation you want, like cube root, etc.
cutTo avoid creating a scale with confusing values, you could use cut to establish sensible category boundaries, and convert these to a numeric scale which is labeled with the original count values:
plot.df.log.cut <- ggplot(data = df, aes(x = x, y = y)) +
geom_hex(aes(fill = stat(cut(log(count), breaks = log(c(0, 1, 2, 4, Inf)), labels = F, right = T, include.lowest = T)))) +
scale_fill_continuous(name = 'count', labels = c('1', '2', '4', '8+'))
print(plot.df.log.cut)
