[GRASS-user] Grouping spatial points

[Johannes Radinger]

Hi all,

I followed the pathway of using R and its hierarchical clustering tools.
For all interested, I came up with following code that creates clusters
with a user-specified minimum distance between the cluster means.
So this code does not specify the number of clusters a priori.

x <- runif(100,0,150)
y <- runif(100,0,150)
df <- data.frame(x,y)
plot(df)

hc <- hclust(dist(df,method = "euclidean"), method="average")
plot(hc)

df$memb <- cutree(hc, h = 60)

cent <- NULL
for(k in 1:length(unique(df$memb))){
  cent <- rbind(cent, colMeans(df[df$memb == k, , drop = FALSE]))
}
cent <- as.data.frame(cent)
summary(dist(cent))

plot(df$x,df$y,col=df$memb)
points(cent$x,cent$y,pch=15)

Of course this could also be used with a distance matrix provided
by any GRASS output (v.distance or distance along a network etc.)

Most probably there are also similar libraries for hierachical clustering
and cutting cluster dendrogramms available for python. Maybe together
with Pietro's approach, its worth considering to write a short add-on, if
this
is also of interest to other users.

Thanks for all your hints!

cheers,
Johannes


On Wed, Jul 9, 2014 at 6:11 PM, Pietro <peter.zamb at gmail.com> wrote:

> Hi Johannes,
>
> On Tue, Jul 8, 2014 at 1:58 PM, Johannes Radinger
> <johannesradinger at gmail.com> wrote:
> > I've a point vector containing more than 500 points.
> > Some of these points are spatially clumped while
> > others are single independent points (from viewing the
> > map). Now I am wondering if there is any tool in
> > GRASS (or maybe other spatial-statistical software
> > like R) that can be used to group the data so that each
> > point clump is assigned to a group and each single group to its
> > own group. Of course, this needs a criterion where
> > to distinguish between groups/clusters. I'd like to have
> > groups that are separated by a distance of at least 5 km.
> >
> > Is there any recommendation of a simple or more
> > advanced procedure to do that?
>
> I play a bit with some machine learning algorithm, the code is:
>
> https://gist.github.com/zarch/f849df450d622f9eadee
>
> you can see the results apply to the vector map "points_of_interest"
> present in the North Carolina mapset, here:
>
> https://plus.google.com/115946523992149326262/posts/2R9HfQh6UvQ
>
> cheers
>
> Pietro
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/grass-user/attachments/20140709/6d783a3e/attachment-0001.html>