I am dealing with a case in which I have many thousands of points in a huge area. As a result, the cost matrix computation takes an exceedingly long time. I think that this could be mitigated if there were an optional parameter which, when specified, will make allneighbordistances quickly compute the euclidean distance between each client/facility pair, and if it is greater than this parameter, will simply set the cost matrix value to NaN. I think that this could really speed up computation in this case, at the expense of losing some connectivity information in the cost matrix - but in some cases it may not be relevant.

I think that this could really help in cases where you are doing facility location optimization with a very large number of candidate sites in a very dense environment (like a city), where the network travel distance is actually based on walking speed. In a case like this, it is quite likely to be irrelevant to compute the walking distance from one end of the city to another.

I would be interested in trying to implement something like this, but I'm new to the library and would like to get opinions & caveats from more experienced people on it.