Closed jschueller closed 1 year ago
Now a real Bonmin developper would be needed :scream:
What you point looks correct as far as I can remember and I can see from the trace. There is a specific method for OA for the Feasibility pump that is called. Then something goes wrong when solving an NLP.
I tried to run it but didn't get that error. On the other hand there's an infinite loop with a quite trivially wrong update to cutoff in iFP.
After fixing it, it's still an infinite loop... but that is numerical and iFP is a bit dangerous. The parameter cutoff_decr
should be set to a larger value for this case (1e-4 works).
@pobonomo Sorry I thought you had left. Great to see that "a real Bonmin developer" is still around :).
@jschueller I cannot reproduce this either, neither with Bonmin 1.8.8, nor stable/1.8, nor current master. You could increase some print or log levels to see why the NLP solve failed. (Maybe your Ipopt isn't working?) A number of ouput options are listed at https://coin-or.github.io/Bonmin/options_list.html
Setting option nlp_failure_behavior could also help to get around this exception (though iFP may then just terminate).
I can confirm the infinite loop that @pobonomo observed, and that the patch from #37 helps to get into a different (nicer looking) infinite loop, and that setting a cutoff_decr
helps to get around that.
yes, that works, thanks a lot
similar to with #31 ifp throws on the simple example with a backtrace similar to: (replace BB with iFP in examples/CppExample/MyBonmin.cpp)
this time the MinlpFeasPump class has no generateCuts method at all (Bonmin::OaDecompositionBase::generateCuts is called), is this allowed ? cc @svigerske @pobonomo