[sdoering]

It should be noted that it is a project from 2015.

But just looking at the clusters, esp. on the east coast and look at the marked known locations one can visually see that the 22 clusters don't make sense. You can visually cluster a lot of stores towards the existing warehouses and remove the need for the estimated ones in between.

But still this is an interesting approach as it helps in identifying the clusters that seem to be served by these known warehouses.

Maybe not taking the distance as the crow flies, but trying a street route could be an optimization. And then trying to find the minimal amount of clusters that still show the known locations at roughly the center.

Not sure if this would be a valid approach.