[postgis-users] 'clustering' of points

[Amit Kulkarni]

Josh,

You might want to look at Crimestat. Ned Levine has created a nice
little program which is *not* crime specific. It may suffice for your
needs, if you dump the stuff into X, Y points. I had emailed Ned a year
back about integrating some scripting functionality and he emailed me
that some people had got it working using VB/VBScript. Unfortunately,
Crimestat is Windows only.

Bill,
My only comment: I sometimes despair at the hacks we have to create in
order to fulfill our time/space constraints. I wish there was a better
way to attack speed/memory problems involved with int/float/double
arithmetic. Hmm, a good topic for some research..

Thanks
-----------------------------------

This is somewhat related to a thread about "density mapping" that took 
place on the mapserver list.

http://thread.gmane.org/gmane.comp.gis.mapserver.user/16860

My last response 
(http://article.gmane.org/gmane.comp.gis.mapserver.user/17924) shows
how 
I accomplished a similar task, and how I would add it into Mapserver.

One thing to note, however, is that this is for visualization, not 
statistics.  If you're looking for "real" kernel density, a tool like R

or GRASS might be better.

Bill

Josh Livni wrote:
> I am creating a map where it would be useful to cluster points, such 
> that if many points were 'close' together, the map instead displays a

> 'cluster' point for that area.
>
> Right now I have a python script that queries my postgis database for

> points with a bbox, and then I crudely break up the bbox into small 
> grids, count the points within it, and if there are lots, I may 
> replace some of those points with 'cluster' point.
>
> But, this is very crude.  I am wondering if there's a clever way to 
> make some kind of SQL query, such that if there are a 'lot' of points

> near a point, it will look at all 'nearby' points, and then return 
> also 'center point' (perhaps a new point that's the average of nearby

> points) along with a list of points included in that 'center points' 
> cluster.
>
> And assuming there's not a pure SQL query, I am guessing this is a 
> problem that people have looked at before, but I don't know in what 
> context or jargon.  So, I hope the above makes sense, and someone has

> an algorithm or better jargon words they can point me to.
>
> Thanks,
>
>   -Josh

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com