Heatmap bbox from clustering

[joetimmerman]

Decided to try tackling #7 by using DBSCAN to cluster the (center points of the) tracks, then only create the heatmap for the first cluster. The code is very messy right now -- just a PoC to see if I could get it work -- and I'm not an expert on clustering algorithms, but it seems to be working well enough for me. Without customizing the heatmap center/radius, my heatmap was gibberish. Adding in the clustering, I get this nice result:

If anyone wants to give it a try, it's in my fork -- I'd love to hear feedback/if it works for data other than mine. I've included a new requirements file, requirements-with-cluster.txt. The only changes are to heatmap_drawer.py, where I've added a new variable _cluster, which can be toggled on or off. You can tweak the eps parameter of DBSCAN to get different cluster sizes -- I'd imagine a cyclist doing 200 mile rides might need this set differently than a runner.

Edit: These changes are now present in this branch of my fork, but nost master.

[flopp]

Looks like your home turf is NYC ;)

I really like the idea of using an automatic clustering algorithm for creating "meaningful" heatmap posters. I will give it a try...