Ilya Builuk

It should be sufficient to make it work, but for bringing it to master branch more steps are needed: - adjust documentation - validation rule considerations: - if zero costs...

That's possible to do, however, technically, solver uses not only cost calculated as linear combination of time and distance, but additionally applies some penalties to guide search in proper direction...

I think the easiest and less surprising way would be to model these costs not like replacement of duration/distance, but rather add it on top. This could be done relatively...

Yes, thread scheduling could be one of explanations: insertion evaluation process is sensible to result comparison order, if results are the same, it will pick the one passed as "right"...

Hi, I think you can try to model this with job and vehicle skills, see documentation and example file: https://reinterpretcat.github.io/vrp/concepts/pragmatic/problem/jobs.html https://github.com/reinterpretcat/vrp/blob/master/examples/data/pragmatic/basics/skills.basic.problem.json

Seems like I missed a job amount constraint from your initial message.. If you have such constraint, then I would agree with @leobudima - multi-dimensional capacity/demand could be used here.

Such problem type directly through its pragmatic format - no. But you can model some of directly unsupported features using built-in features, like one mentioned above. This is why api...

Hm, it seems a side effect of this change (I'm using github action to build them automatically): https://github.com/reinterpretcat/vrp/commit/397d5a9965e06c3add73565e6482640b8d74597b I did it to workaround an issue I got with old code:...

Hi, thanks for feedback! Looks like an interesting issue, never seen it so far. Seems some solutions are added to the population in an invalid state: either `init_weights` is not...

a new version (1.22.0) should support this when a specific option is set in the config. See https://reinterpretcat.github.io/vrp/getting-started/solver.html#geojson for more details