I'm quite surprised this went past the radar, because it's very interesting research and abuses openstreetmap-carto style in a novel way:
https://www.sentiance.com/2018/05/03/loc2vec-learning-location-embeddings-w-triplet-loss-networks/ Feed in openstreetmap-carto rendered in 12 thematic layers to a neural network and get a 16-dimensional vector. This vector captures nature of the place (e.g. urban vs rural, city center vs suburb etc.) Source is not avaliable, but many details are present, so somebody could recreate it or get inspired by this work. Michał
_______________________________________________ dev mailing list [email protected] https://lists.openstreetmap.org/listinfo/dev
