blogle
commented
9 years ago Turns out the most obvious cause is that initial insertion of edges into the priority queue is based on the sq_dist instead of haversine for performance optimization. Before simply changing this to haversine in the inner loop, perhaps another cost effective methodology could be employed here. While this resolves many errors Boruvka is inherently slightly different from Kruskal.
chrisnatali
I need to go through the core algo to see if there is anything that is obviously causing this error. Furthermore, might be worth considering parallelizing the nearest neighbors search while in there.