optional activation

[apete]

@Programmer-Magnus

You mentioned once that you had a problem with the QP solver initiating itself with a lot of inequalities active, and then spending time deactivating those, to eventually return a solution with no or very few inequalities active. I assume this is when using extendedPrecision and the IterativeRefinementSolver. Can you test if this small change makes a difference in your case?

[Programmer-Magnus]

The QP branch made it better. :)
I did not manage to find the used number of iterations in the result object. I just now realized, maybe it is accessible in the model or elsewhere, after solving EBM-models?
I enabled logging and run a set of models (abut 7000 of them) . Then parsed the logs and summed the internal iterations. It used 47757 iterations with develop, 39785 iterations using QP branch, that is 83% of the former.
Previously I probably run one ore two ojalgo models, maybe the ones I tried showed extreme behavior.

[apete]

17% "better" is significant. So, do think this is preferable behaviour when using the IterativeRefinementSolver? Is there some rationale for why this would be a better configuration?

Actually, all iterations are not equal. With that many models (I assume you run them using some sort of script/code) why don't you just measure the time it takes to complete the whole set (with logging turned off). If there are different groups of models (size, problem type) perhaps try them separately.