Debuging on YALMIP

54 views
Skip to first unread message

othman moumni abdou

unread,
Sep 6, 2013, 3:58:29 AM9/6/13
to yal...@googlegroups.com
Hi,
I am using YALMIP to model my problem (LP) and solve it using Cplex.

When I run solvesdp, I obtain the following error:
Primal unbounded due to dual bounds, variable 'x3'.

The problem is that I don't know which of my sdpvar variables are related to 'x3'.
Is there a way to show the YALMIP computed model that is sent to Cplex? or just a way to know to which variable 'x3' is related to in my model?

Thanks in advance

Othman

Johan Löfberg

unread,
Sep 6, 2013, 4:01:13 AM9/6/13
to yal...@googlegroups.com
You can use the command export to get the numerical model, but it will still be tricky to figure out how the indexing refers to your symbolic model (but simply speaking, x3 should be the third variable you have defined (and if the first variable you defined was of length 5, it thus means that x3 is the third element in that vector)

Perhaps easier if you post the model here, sometimes it is trivial to see

othman moumni abdou

unread,
Sep 6, 2013, 4:58:12 AM9/6/13
to yal...@googlegroups.com
Ok, thank you for you fast answering.

I export the model to an AMPL format (see attached file).

Othman

myamplmodel.mod

Johan Löfberg

unread,
Sep 6, 2013, 5:08:34 AM9/6/13
to yal...@googlegroups.com
I want the YALMIP model of course
Message has been deleted

Johan Löfberg

unread,
Sep 6, 2013, 7:00:51 AM9/6/13
to yal...@googlegroups.com
Obviously something is wrong with the model since I can make the objective arbitrarily large by letting variables tend to infinity in various directions

sol = solvesdp([constraints,recover(depends(objective)) <= 100000000],objective)

>> p_stockneg
Linear scalar (real, 1 variable, current value : -99999975.7009)
>> p_serviceAchat
Linear scalar (real, 1 variable, current value : -78571452.8706)
>> p_serviceVente
Linear scalar (real, 1 variable, current value : 100000000)


othman moumni abdou

unread,
Sep 6, 2013, 7:09:12 AM9/6/13
to yal...@googlegroups.com
ok, I will search in this direction.

Thank you
Reply all
Reply to author
Forward
0 new messages