coin-or
Setting `primal_feasibility_tolerance` for `pulp.HiGHS` solving
What is your question
I am solving a problem via pulp.HiGHS but consistently get a "sub-optimal" solution with the following errors:
ERROR: MIP solver claims optimality, but with num/max/sum primal(1/2.38419e-06/2.38419e-06) infeasibilities
ERROR: Setting model status to Solve error
It is only ever a super tiny violation which presumably is some mathematical artefact (I hope?) and so I would rather not false-trigger the switch of model status as I am running a huge grid and programmatically interface with summaries of the results. I tried setting primal_feasibility_tolerance in many ways (found in HiGHS API) but cannot seem to get it to stick; solverParams from the pulp docs doesn't appear to do anything and in fact causes an error if I pass a dict into it.
Any help would be appreciated, might be that I am missing some docs somewhere!
I suspect you have number instability in your model (too many decimals, or too large of a number). Trying rounding values to a reasonable degree (e.g., integers if possible) and see if that solves the issue
I think you are correct, I rescaled the objective using the initial total of one of the components and it has disappeared, thank you!
The issue still stands I think as it does seem impossible to tweak this parameter, but less urgent for me anyway!