Convex hulls in scatter plots

[KDB]

Hi everybody,

I was wondering if there is an elegant way to produce convex hulls for
several groups in scatterplots created with the ggplot2 package. I
found this post by learnr very helpful:
http://learnr.wordpress.com/2009/04/23/ggplot2-dont-try-this-with-excel/
The use of the geom_path() or geom_polygon() seem logical to
accomplish this, but the solution proposed in this blog doesn't seem
to work in my case. I have a quite large dataset and sometimes point
are connected in the way they are order in the dataset. I know several
packages who offer functions to calculate convex hulls for x and y
data (e.g. chull() ) in package grDevices). Is there some easy way to
integrate their use with ggplot ??? I guess if we can somehow select
the points from the main dataset who lie on the convex hull as
calculated by chull() or a similar functions and connect them with
lines (geom_path) or polygons (geom_polygon) shoudl do the trick.

I am a beginner in R and found ggplot quite straight forward and easy
to use, but i don't know how to do this.
Your help would be greatly appreciated.

Kenneth

[Hadley Wickham]

Hi Kenneth,

The basic idea is something like:

library(ggplot2)

df <- data.frame(x = rnorm(100), y = rnorm(100))
ggplot(df, aes(x, y)) + geom_point()

hull <- df[chull(df$x, df$y), ]
ggplot(df, aes(x, y)) +
geom_point() +
geom_polygon(data = hull, fill = NA, colour = "grey50")

If you have multiple groups, you'd need to use ddply or similar to
compute the convex hull for each group.

Hadley

> --
> 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
>

--
Assistant Professor / Dobelman Family Junior Chair
Department of Statistics / Rice University
http://had.co.nz/

[KDB]

Hi Hadley,

Thanks for your help as i am now capable of making convex hulls for
the entire dataset.
I still haven't managed to create it for groups. Would it be possible
to give an example of the use of ddply in this context.

With friendly greetings,

Kenneth

[Hadley Wickham]

Hi Kenneth,

Something like this:

library(ggplot2)

df <- data.frame(x = rnorm(100), y = rnorm(100),
z = sample(letters[1:5], 100, rep = T))
ggplot(df, aes(x, y, colour = z)) + geom_point()

find_hull <- function(df) df[chull(df$x, df$y), ]
hulls <- ddply(df, "z", find_hull)

ggplot(df, aes(x, y, colour = z)) +
geom_point() +
geom_polygon(data = hulls, fill = NA)

But be careful with the use of hulls. The tend to make clusters look
more distinct than they really are (due to the gestalt of
connectedness/containedness)

df$cl <- factor(kmeans(df[c("x", "y")], 3)$cluster)
hulls <- ddply(df, "cl", find_hull)

# No obvious clusters
ggplot(df, aes(x, y, colour = cl)) +
geom_point()

# Now clusters are really obvious!!
ggplot(df, aes(x, y, colour = cl)) +
geom_point() +
geom_polygon(data = hulls, fill = NA)

Hadley

http://had.co.nz/

[KDB]

Thanks Hadley, it work like a charm.

One additional question: i seem to run into trouble, when i want to
give the points in my scatterplot a shape according to one parameters
and the convex hulls to group points reflect another parameter ?
How can this be avoided ?

On Mar 14, 2:10 pm, Hadley Wickham <had...@rice.edu> wrote:
> Hi Kenneth,
>