generate unique colors for categorical color argument to plotSPC

[dylanbeaudette]

plotSPC does not generate enough colors in the legend (?) when coloring horizons with categorical data containing more than 8 unique values.

[pierreroudier]

This is a limitation of RColorBrewer isn't it?

A more flexible method could be to distribute the N colours evenly on a HSV circle. I think this is what ggplot2 is doing by default. This may be facilitated by the colorspace package by the looks of things.

[pierreroudier]

Another solution would be random colours. Some good pointers:

https://stackoverflow.com/questions/15282580/how-to-generate-a-number-of-most-distinctive-colors-in-r
http://martin.ankerl.com/2009/12/09/how-to-create-random-colors-programmatically/

[dylanbeaudette]

Good idea. I suppose than more than 8-10 classes is not our responsibility as the legend becomes impossible to use. Random colors would be nice so as not to convey some kind of ordering when there is none.

Several possible paths:

1. character class, use random colors
2. factor class, use default ordered colors after expanding with colorRampPalette
3. in the case of >8-10 classes, issue a warning

There is still a problem with the legend: the current implementation doesn't scale well to > 8 classes.

[dylanbeaudette]

I think that this issue is now 90% solved:

• colors are mapped to categorical data, even when n > length(col.palette)
• the legend is split into 2 rows when n > n.legend

Try it out.

library(aqp)

data(sp3)
depths(sp3) <- id ~ top + bottom

# make some fake categorical data
horizons(sp3)$fake.data <- sample(letters[1:15], size = nrow(sp3), replace=TRUE)

# better margins
par(mar=c(0,0,3,1))

# note that there are enough colors for 15 classes (vs. previous limit of 10)
# note that the legend is split into 2 rows when length(classes) > n.legend argument
plot(sp3, color='fake.data', name='fake.data', cex.names=0.8)

# make enough room in a single legend row
plot(sp3, color='fake.data', name='fake.data', cex.names=0.8, n.legend=15)