[QST] How to find out which primal heuristics were applied

[shariharan2002]

Hello!
So, if I run cuopt_cli on an mps instance - is there a way to:

1. Apply only MIP primal heuristics
2. Apply only MIP primal heuristics that don't involve solving SubMIPs(eg: disable LNS/RINS)
3. Actually see which MIP primal heuristics found the best/first solution?
   What are the flags and can you give me a sample command or tell me where should I look for this in the documentation?

Thanks

[rgsl888prabhu]

Hello! So, if I run cuopt_cli on an mps instance - is there a way to:

1. Apply only MIP primal heuristics
   Yes, use the --mip-heuristics-only, This runs only the GPU-based primal heuristics (Feasibility Jump, genetic recombination, local search, etc.) and disables the CPU-based Branch & Bound tree search. Only the primal bound is improved; no dual bound is computed.
2. Apply only MIP primal heuristics that don't involve solving SubMIPs(eg: disable LNS/RINS)
   Unfortunately, there's currently no CLI flag to disable RINS specifically.
3. Actually see which MIP primal heuristics found the best/first solution?> What are the flags and can you give me a sample command or tell me where should I look for this in the documentation?>
   This can be observed if we enable debug log level, but we currently don't expose debug log level.

@chris-maes Please correct me if I am wrong.

[shariharan2002]

I see! Thanks for the detailed response @rgsl888prabhu
So, suppose, I build cu-opt from source and then say add a few logging print statements around feasibility jump and then say even go on to disable RINS -- is that possible to do so? Or is that practically unachievable for a person who has seen the cuopt code just for a week or so?

[rgsl888prabhu]

I see! Thanks for the detailed response @rgsl888prabhu So, suppose, I build cu-opt from source and then say add a few logging print statements around feasibility jump and then say even go on to disable RINS -- is that possible to do so? Or is that practically unachievable for a person who has seen the cuopt code just for a week or so?

With coding agents help you, you might be able to to pull it off. But I would wait for @chris-maes input on feasibility and any unforeseen behavior which might not yield the results you expect.

[aliceb-nv]

I think disabling RINS should just be a matter of commenting out this line:

cuopt/cpp/src/mip/diversity/diversity_manager.cu

Line 473 in 514daf9

rins.enable();

[shariharan2002]

Great!
@aliceb-nv @rgsl888prabhu - just wanted to confirm - is RINS the only sub-MIP heuristic implemented in cuOpt set of primal heuristics as of now? Or are there more? I can only see RINS in the LNS directory. I am not sure if recombiners are part of sub-MIP.

If so, then I guess I just need to comment out as @aliceb-nv told the rins.enable(); line and then add print statements wherever needed to just track which heuristic is actually giving solutions for the problems that I have!
Seems doable.

[akifcorduk]

In the solver, there is an option called heuristic_only https://github.com/NVIDIA/cuopt/blob/main/cpp/include/cuopt/linear_programming/mip/solver_settings.hpp#L87. That prevents full mip solver from runing. But the heuristics themselves use full mip solver as a subprocedure in sub_mip crossover. If you also disable that, you might achieve fully heuristic results.

[github-actions]

🔔 Hi @anandhkb @rgsl888prabhu, this issue has had no activity for 21 days. Please update or let us know if it can be closed. Thank you!

If this is an "epic" issue, then please add the "epic" label to this issue.
If it is a PR and not ready for review, then please convert this to draft.
If you just want to switch off this notification, then use the "skip inactivity reminder" label.

[github-actions]

🔔 Hi @anandhkb @rgsl888prabhu, this issue has had no activity for 21 days. Please update or let us know if it can be closed. Thank you!

If this is an "epic" issue, then please add the "epic" label to this issue.
If it is a PR and not ready for review, then please convert this to draft.
If you just want to switch off this notification, then use the "skip inactivity reminder" label.

[github-actions]

🔔 Hi @anandhkb @rgsl888prabhu, this issue has had no activity for 21 days. Please update or let us know if it can be closed. Thank you!

If this is an "epic" issue, then please add the "epic" label to this issue.
If it is a PR and not ready for review, then please convert this to draft.
If you just want to switch off this notification, then use the "skip inactivity reminder" label.