I did a project where we produced maps of car dealerships. We clustered close together dealerships. The distance function that I used for measuring "closeness"

I did a project where we produced maps of car dealerships. We clustered close together dealerships. The distance function that I used for measuring "closeness" evolved from a simple Euclidean distance (sqrt(∆x²+∆y²)) to one that used box-shaped neighborhoods (max(∆x,∆y)) to a variation on this last one that took Mercator distortions into account because Alaska. This was one of the most fun projects I worked on. There was one subtle bug that I had in the web interface where at the US level, clicking on the wrong place in the spash screen loaded, instead of the clustered dealerships at that level, instead added a pin to the map for every single dealership in the country. It took about 10 minutes to render on 2007 hardware but was really pretty when it was done. I no longer have the screenshot unfortunately.