sampled data, plain distance function

Here's the first result i've produced. The data is sampled 1/100 for performance reasons. There are 11 districts. Cost function used is distance function to the cluster point. 8 iterations of k-means were used.

The districts look "shrink-wrapped" because i used sampling and the concave hull operation.

I've learned a few things from these results.

1. Using normal distance function in the k-mean function doesn't make sense, when it produces a cluster that crosses the Chesapeake Bay.
2. This obviously needs refinement, because it splits Richmond into 2 parts.