Point Clustering Utility Trigger

[pramsey]

Generating cluster markers for web user interfaces is v. v. common. Function(s) that can run on insert/update/delete that manage a '*_clustered' table would allow people to delegate that problem to the database.

[pimpaa]

Replying to pramsey:

Generating cluster markers for web user interfaces is v. v. common. Function(s) that can run on insert/update/delete that manage a '*_clustered' table would allow people to delegate that problem to the database.

Is this to be used with SnapToGrid? If so i think i can try something for this, altought i´ll wait and finish the two other tickets...

George

[pramsey]

No, this is a good deal more involved, potentially. k-means clustering perhaps. gridded systems maybe. it needs some thought, to balance out quality of clustering with cpu load.

[brian252]

I've tried SnapToGrid in the past and it doesn't look impressive.

I wrote up an implementation of k-means clustering. It's not perfect, but it will work for web-maps. It clustered 22,000 points (tiger.tl_2009_51_cty.pointlm) in about 30 seconds.

The algorithm isn't complicated, but i'm not sure how to set up the mechanics of creating this auxiliary table. It currently runs for n iterations. It would be better to run the clustering until the clusters move less than some delta.

It's hosted here ​http://brian.derocher.org/?Page=project&Id=104201

[strk]

An incremental approach is described here: ​http://web.eecs.utk.edu/~itamar/Papers/ITNG2010.pdf

It would help with maintenance of the '*_clustered' idea

[pramsey]

Clustering functions are being added, DBScan (#3362) and KMeans (#3465).

[robe]

Milestone renamed