Open svigerske opened 5 years ago
Attachment minlp_trouble.osil by @tkelman created at 2013-10-18 19:47:02
maximization version of the problem
Attachment minlp_works.osil by @tkelman created at 2013-10-18 19:47:25
minimization version of the problem
Attachment options.osol by @tkelman created at 2013-10-18 19:48:02
some useful options
Comment by JunMa created at 2015-03-13 17:52:34
Sorry for the late reply. Do you still encounter the same issue?
Comment by @tkelman created at 2015-04-14 07:41:36
Yes.
Issue created by migration from Trac.
Original creator: @tkelman
Original creation time: 2013-10-18 19:46:33
Version:
Keywords: bonmin maximization minlp
Examples attached: minlp_works.osil is a minimization version of the problem, minlp_trouble.osil is a maximization. As you can see by setting nlp_log_at_root to something normal like 5, the continuous relaxation at the root node doesn't convergence properly. The derivative checker indicates the objective Hessian has the wrong sign.
It looks like the Bonmin interface doesn't do the same objective scaling that the Ipopt interface does for maximization problems. Solving the continuous relaxation directly by changing all the integer variables to continuous works fine for either maximization or minimization.