ggplot2/colorbrewer qualitative pallette with 125 categories

Question

I have data as follows:

• 10 states
• Each state has two types
• Each type has between 1 and 29 entities
• Each state-entity-type has a count

Complete data available as a gist.

I'm trying to visualize what proportion of the counts were made for each entity. To do that, I've used the following code:

icc <- transform( icc, state=factor(state), entity=factor(entity), type=factor(type) )
p <- ggplot( icc, aes( x=state, y=count, fill=entity ) ) +
  geom_bar( stat="identity", position="stack" ) +
  facet_grid( type ~ . )
custom_theme <- theme_update(legend.position="none")
p

Unfortunately, I'm losing a lot of information because state-types with lots of entities aren't displaying enough unique colors.

As mentioned above, I have 125 entities, but the most entities in a state-type is 29. Is there a way to force ggplot2 and colorbrewer to assign a unique (and hopefully fairly distinct) color within each entity-type?

The only way I've come up with so far is to coerce entity to an integer, which works but doesn't provide much color differentiation between levels.

Northrop

@Arun Producing such a pallette with a nice diverging scheme is non-trivial (might be a good answer, wink wink). Need to run a color brewer on the entities each state-type (itself problematic, since most pallettes choke on more than about 12, it seems), then put it in the right order. – Northrop

It's very difficult to come up with 29 distinct colours. Why not just label the entities with text? – Stinky

@Stinky Noted. Text would be great, but is likely to end up a jumbled mess. I guess my requirement is actually much simpler: the colors don't even have to be unique within a state-type, just alternating. I can try something like scale_fill_manual( values=c(rep(c('red','blue'),125/2),'red') ), but then they're not guaranteed to be alternating within each bar. – Northrop

You could probably create a new variable with alternate <- function(x) factor(match(x, unique(x)) %% 2)), and apply it within groups with ave: ave(entity, state, FUN = alternate) (untested) – Stinky

Answer 1

Here's an approach that gives you a little more information. Take the color wheel generated by rainbow, and for every other color, swap it with the opposite one on the wheel.

col <- rainbow(30)
col.index <- ifelse(seq(col) %% 2, 
                    seq(col), 
                    (seq(ceiling(length(col)/2), length.out=length(col)) %% length(col)) + 1)
mixed <- col[col.index]

p <- ggplot(icc, aes(x=state, y=count, fill=entity)) +
  geom_bar(stat="identity", position="stack") +
  facet_grid( type ~ . ) + 
  scale_fill_manual(values=rep(mixed, length.out=nrow(icc)))

custom_theme <- theme_update(legend.position='none')
p

Lots