Problem using optimizer with Octave

[Tobias L]

Hi Johan,

I tried to use the optimizer command to speedup many optimizations with different inputs to the objective.

Unfortunately it seems that the optimizer command behaves strange in octave, but works fine in Matlab.

Consider the following simple example, tested with Octave (4.0 Linux / 4.2.1 Windows).

min a*x

s.t. 1 <= x <= 10

x=sdpvar; a=sdpvar;

p=optimizer([x>=1, x<=10], a*x', sdpsettings('solver', 'glpk'), a, x);

[l,f]=p(-1)

>> l = 1 

It gets even more strange when using the solver sdpt3, then I get

>> l = 5.5000

for a = 1 and a = -1

However, using optimize gives the correct solution (as does Matlab with the above and solver=linprog):

optimize([x>=1, x<=10], -1*x', sdpsettings('solver', 'glpk')) % or sdpt3

l = value(x)

>> l = 10

Do you have any clue what is causing this problem and is there a chance to fix it?

Thank you and best regards
Tobias

[Johan Löfberg]

I will try to get Octave installed and see, but it will not be done in a couple of days

Your best and fastest bet is to simply put a break in callglpkoctave (breaks-points possible in octave?...) and see if the correct data is sent to the solver, and then debug from there

[Tobias L]

Thanks for the quick reply.

After stepping through callglpkoctave I can confirm that the data gets corrupted when using the optimizer command.

Instead of A=[-1 0; 1 0]; B=[-1; 10]; C=-1 as when directly calling optimize,

it gets A=[-1 0; 1 0; 0 -1; 0 -4]; B=[-1; 10; 1; 0]; C=[0 1];

Thank you for your time.

[Johan Löfberg]

Add R = []; on line 444 in lmi/categorizeproblem

[Tobias L]

Great, it works now.

Thank you for the quick fix!