Re: [Bug-glpk] glp_intopt recovery feature added

[Andrew Makhorin]

Hi Chris,

> > Currently the solver makes up to 5 attempts to choose an
> > appropriate column, i.e. having not too small pivot, but the pivot
> > magnitude is not checked.
> 
> I had the first patch attached in my tree to give errors instead of
> the assert. Following this discussion I did a quick test (seen in the
> second patch) to try a new pivot with a simple threshold. I'm not sure
> whether this is the best option, but it seems to help in several tests
> I did.

Thank you.

I replaced choose_pivot (an internal routine, which is a crucial point)
in both primal and dual simplex solvers, and now glpsol works much more
stable (in particular, it is able to process water.mps as well as some
other numerically hard lp and mip models). However, the features I added
are highly experimental, though I hope there must be no degradation in
the solvers. Will try to make a pre-release.

Best regards,

Andrew Makhorin