re: Labeled proportional symbols: How'd they do that?

pauladamsmith — Sat, 06 Sep 2008 17:09:48 GMT

Our algorithm for clustering the features is actually independent of the underlying areal geometries. The way it works is similar to k-means clustering, except instead of creating k number of clusters, we use a radius to determine whether features should be clustered together. For each feature, we ask if it falls within a buffer defined by the radius and if so, we add it to the cluster, otherwise, we create a new cluster (we initialize by arbitrarily assigning the first feature to the first cluster). The centroid of a cluster is updated with each new feature by simply finding the mean of the component features' centroids.

For EveryBlock, the value of the radius we choose for the clustering function is actually just the size of the largest of the 5 cluster icons we use (in pixels but obviously translated into units of the map's spatial reference system). As you zoom we re-cluster based on the implicit change to the radius due to the new map scale.

I'm not an ArcGIS user so I'm not sure if any of this is useful but I thought I'd share how we went about it. Thanks for highlighting this feature!

re: Labeled proportional symbols: How'd they do that?

gcriloux — Mon, 08 Sep 2008 21:41:50 GMT

Thanks for both the article and Paul's remark.

Isn't k-mean clustering sensitive to how you seed the algorithm ? how do you pick the seeds ?