Suggestion/Wish for additional features

[OptimusCrime]

Hello!

We are using your excellent library and we have implemented a few additional features that you may be interested in adapting. These features includes:

• Requested time - Minimize deviation from requested service times
• Ride duration - Limits the ride duration for pickup-delivery jobs
• Solo riding - A constraint for people that can not ride with other passengers, even if the vehicle has capacity
• Preferences - A soft constraint that allow us to express preferred vehicles without making them hard requirements
• LIFO ordering - A constraint that ensures that passengers are loaded and unloaded in the correct order

We need all of these features to be able to express real world modelling for our use-case. We have these features implemented in our fork, and we would be happy to provide PRs to get them into the core library, if you find them relevant and usable.

[reinterpretcat]

Hello!

I'm thinking that it might sense to integrate those once I release a new version from master branch (it will take some time, need to verify few things). I might play with some extra changes (mostly on algorithmic/performance side burning my LLM subscription tokens trying to find potential improvements), but the code is basically ready.

Regarding contributions, typically, I pay attention in the new code on:

• overall consistency of design
• impact on other features
• as pragmatic format most likely is affected, the way feature is modeled on API should be also generalized well and consistent with the rest

So, basically my approach is attempt to maintain conceptual integrity across the repo.

[OptimusCrime]

Thank you. That makes a lot of sense. One developer on our team has done this over the last couple of months (letting the LLM find improvements and optimizations). He had some initial trouble with the library when he tried to solve big datasets. I can not really speak for the improvements myself, as I have not really looked into what they do, but you could take a look here if you are interested: https://github.com/henningms/vrp

I see that your branch has been merged. We will discuss internally how we proceed with the additional constraints that we need. For now we have pointed our application to a fork, so there is no rush.

[shubham441996]

Hi @reinterpretcat ,

Is there a changelog, draft release note, or documentation available that summarizes the recent changes merged into master?
Also, in my initial benchmarks, master shows lower gen/sec and lower CPU utilization compared to the previous master (before vnext_next merge) . Is the solver performing heavier computations or deeper local searches per iteration now to converge on better overall solutions, or have there been changes to thread/CPU scheduling?

[reinterpretcat]

I still need to update change log, but basically I did quite a lot of changes on improving (I believe) algorithm. I think it is not comparable by gen/sec with old version. So, better to measure speed of convergence, e.g. how fast you get better solutions over time.

Also there are more initial solutions built initially which also might take a bit longer (but not more than 5% time quota from max-time when specified).

lower CPU utilization

this is concerning, yes. If results are quite worse than before, maybe you can share anonymized problem formulation with matrix file? You can use index variant instead of lat/lng to avoid sharing real coordinates.