I have some thousands of houses/addresses (points) and a road network and I would like to build clusters of defined minimum to maximum cluster size not using euclidean distance as a metric, but the distance between the addresses along the road network (where I'm not sure if I'm using the word 'cluster' in the correct context here).
I read this paper http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.92.6037&rep=rep1&type=pdf and had a look at PostGIS function
ST_ClusterDBSCAN, which is unsuitable for my task because DBSCAN algorithm is not intented to limit cluster size nor is it possible to define a non-euclidean metric for the input data.
K-Means approach might not be suitable, because the numbers of clusters is not defined a-priori due to varying count of addresses within a cluster.
Everything seems to point me to hierachical clustering (I had alook at How to cluster points into clusters of a maximum diameter in PostGIS?), but by now I could not find a solution taking the metric along a spatial network into account.
For better understanding ref. to https://youtu.be/iQvpPl-J7b4?t=328 , the guys do something similar to what I'm after.
Are there any open solutions to this?