Hi all, I am now trying BONMIN on a MINLP, chose to began with B&B just for the fun of "watching" the tree grow . I don't understand the following process I'm seeing in the solver log:
As you can see there seems to be reports of two different kinds NLPs being run, which I marked in italics and bold respectively.
Cbc0010I After 40 nodes, 2 on tree, 0.0089116746 best solution, best possible 0.00011327011 (1955.42 seconds)
NLP0014I 390 OPT 0.033555868 28 3.34
Cbc0010I After 41 nodes, 2 on tree, 0.0089116746 best solution, best possible 0.00011327011 (1958.76 seconds)
NLP0014I 391 OPT 0.00012181808 36 3.816
Cbc0015I Node 42 Obj 0.00012181808 Unsat 167 depth 7
Cbc0010I After 42 nodes, 2 on tree, 0.0089116746 best solution, best possible 0.00011327011 (1962.58 seconds)
NLP0014I 392 OPT 0.032711641 43 4.538
Cbc0010I After 43 nodes, 2 on tree, 0.0089116746 best solution, best possible 0.00011327011 (1967.12 seconds)
NLP0014I 393 OPT 0.00012379623 47 4.459
Cbc0015I Node 44 Obj 0.00012379623 Unsat 166 depth 8
Cbc0010I After 44 nodes, 2 on tree, 0.0089116746 best solution, best possible 0.00011327011 (1971.58 seconds)
The one in italic corresponds clearly to the succesive branching, this can be seen since its objective function value ends up in the B&B log (Node xxxx Obj xxxx Unsat xxxx Depth).
But, what are the other NLPs, which generated outputs marked in bold, being run ? Is it another node of the tree not on the current branch ?
On the same topic: Before starting the B&B process, BONMIN ran a series of NLPs, the number matching with twice the number of binary variables. What are those, too ?
Thanks in advance
Solver related questions
3 posts • Page 1 of 1
Not sure how many people understand the BONMIN log. My colleague Stefan Vigerske who implemented the GAMS/BONMIN link says he doesn't understand the log either (he suspects that these NLPs are related to some primal heuristics). I guess you need to go to the computer (https://projects.coin-or.org/Bonmin) or human (http://researcher.ibm.com/researcher/vi ... rre.bonami) source.
Thanks Michael, indeed most solvers have quite obscure logs. I would have to do the research but it seemed worth giving the new forums a shot for anyone familiarized with the math behind BONMIN. Nice to see gamsworld back, btw !