Passing a continuous fill variable to geom_dotplot

[Tom W]

I've been really enjoying the new geoms in 0.9. I've hit an unexpected barrier with geom_dotplot.

d2 <- data.frame(v1 = runif(10, 1, 40),

                 v2 = runif(10, 0,1))

ggplot(d2, aes(x = v1, fill = v2)) + geom_dotplot()

provides the following error message

stat_bindot: binwidth defaulted to range/30. Use 'binwidth = x' to adjust this.
Error in xyz %*% solve(M) : non-conformable arguments

Removing the fill option provides the plot with no errors.

Am I missing something?

Tom

[Jean-Olivier Irisson]

I've had this error quite often also, without being able to track it down. I can say that it's not related to geom_dotplot or stat_bin (which is what emits the first message). I got it with colour of points on a scatter plot and on fill of tiles on a tile plot.

Any insight is welcome.

Jean-Olivier Irisson
---
Observatoire Océanologique
Station Zoologique, B.P. 28, Chemin du Lazaret
06230 Villefranche-sur-Mer
Tel: +33 04 93 76 38 04
Mob: +33 06 21 05 19 90
http://jo.irisson.com/

[Winston Chang]

I can't explain offhand the details of why it gives that error, but I can say that the way geom_dotplot is written, it _shouldn't_ work with a continuous fill.

It uses stat_bindot, which is similar to stat_bin -- essentially, it gets an input data frame, and then returns a data frame with an x value for each bin, and a count of the number of dots in the bin. The individual data points aren't in the new data frame; the new data frame only has counts. Then the geom_dotplot draws a stack of n dots for each bin. In other words, instead of "moving" the data points into each bin, it simply gets a count of the number of cases in each bin.

You get the same error if you try to use geom_histogram, which uses stat_bin.

d2 <- data.frame(v1 = runif(10, 1, 40),

                 v2 = runif(10, 0,1))

ggplot(d2, aes(x = v1, fill = v2)) + geom_dotplot()

# stat_bin: binwidth defaulted to range/30. Use 'binwidth = x' to adjust this.

# Error in xyz %*% solve(M) : non-conformable arguments

ggplot(d2, aes(x = v1, fill = v2)) + geom_histogram()

# stat_bin: binwidth defaulted to range/30. Use 'binwidth = x' to adjust this.

# Error in xyz %*% solve(M) : non-conformable arguments

-Winston

--
You received this message because you are subscribed to the ggplot2 mailing list.
Please provide a reproducible example: http://gist.github.com/270442

To post: email ggp...@googlegroups.com
To unsubscribe: email ggplot2+u...@googlegroups.com
More options: http://groups.google.com/group/ggplot2

[Tom W]

hi Winston--thanks for such an authoritative answer. I appreciate if there's a fundamental technical obstacle to implementing this option.

I also think it would occasionally be useful to have this capability--I'm not sure that it would contradict the underlying grammar. But anyway, thanks so much for the work you've done implementing this new facility--the new features in 0.9 have been tremendous.

Tom

-Winston

To unsubscribe: email ggplot2+unsubscribe@googlegroups.com
More options: http://groups.google.com/group/ggplot2

[Winston Chang]

I'm glad you find it useful!

It occurred to me that it's possible to simulate what you're looking for if you're willing to have a discretized scale. (I think it's also perceptually easier to map colors back to numerical values using a discrete scale.) This requires a development branch of ggplot2 from github, where I added the ability to use stacking with dot plots.

library(devtools)

install_github('ggplot2', 'wch', 'dotplot-stack')

library(ggplot2)

set.seed(12)

d2 <- data.frame(v1 = runif(40, 1, 40),

                 v2 = runif(40, 0, 1))

# Convert the continuous values to discrete

d2$v3 <- cut(d2$v2, seq(0, 1, 0.2))

# Bin positions determined by data (dot-density)

ggplot(d2, aes(x = v1, fill = v3)) +

  geom_dotplot(stackgroups=TRUE, binpositions="all") +

  scale_fill_brewer()

# Regularly-spaced bins

ggplot(d2, aes(x = v1, fill = v3)) +

  geom_dotplot(stackgroups=TRUE, method="histodot") +

  scale_fill_brewer()

Make sure you have the latest devtools because there was a version that had problems installing branches from github.

I'm not sure if building in the capability to use a continuous fill would necessarily violate the grammar... I think it would require a complete reworking of the way stat_bindot works, and how geom_dotplot deals with the binned data. There's also the issue of how to vertically order the dots based on a continuous variable. I can't think of another case where something like this is done, so it would probably have to be a special case written into geom_dotplot. Overall, it probably makes more sense to just use the discrete method with stacking. Hopefully stacking will be in ggplot2 0.9.1.

-Winston

[Winston Chang]

I forgot to mention that you should probably do dev_mode() right after loading devtools, so that you don't mess up your regular installation of ggplot2:

  library(devtools)

  dev_mode()

[Tom W]

wow, thanks! I'll try that our right away.

we users of open source software have no right to expect such responsiveness from volunteer programmers!