OCTAVE (YALMIP -> SeDuMi)

[Alexandre Felipe]

I've tried some times in the last year to get octave running YALMIP for solving SDPs,

Today I read about SPDT3 and installed it however yalmip expected versions 3.0, 3.02 or 3.1, the version currently available is 4.0. The authors seems not to keep the previos versions, so I will try to find it from. 

Fortunately this new release of yalmip interfaces with SDPT3-4.0,

And Now at first time I could solve a SDP in octave :D

Earlier I attempted again, in the hope that the things could go better with the newest release.

I am using octave 3.8.2

I installed sedumi, comented a line from their installation script 'install_sedumi' in which they write a define mwSignedIndex

% flags{end+1} = '-DmwSignedIndex=mwIndex';

added all those folders to the octave path, and tried just as I made in the MATLAB twice, to run "yalmiptest".

In several problems I read the folllowing message "Unknown problem in solver"

What I always inferred from this message was that SeDuMi was the source of the problem,

With the error message there is a hint "Turn on 'debug' in sdpsettings"

After trying a symple problem

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

A = randn(4);

A = A - (max(real(eig(A))) + 0.1)*eye(4);

P = sdpvar(4,4,'symmetric');

LMIs = ((A'*P + P*A <= 0) + (P >= 0));

sol = solvesdp(LMIs, [], sdpsettings('solver', 'sedumi', 'debug', 1));

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

It raised the following error

octave:51> solvesdp(LMI, [], sdpsettings('debug', 1))
error: compileinterfacedata: A(I): index out of bounds; value 27 out of bound 17
error: called from:
error:   .../yalmip/extras/compileinterfacedata.m at line 204, column 16
error:   .../yalmip/solvesdp.m at line 219, column 77

I suspect this is not a difficult bug for those who know the yalmip engine.

I would appreciate a lot an eventual solution.

And I am happy of having some solver working, now it's time to see how well it performs, since I have used only SeDuMi.

[Johan Löfberg]

Not sure what you are talking about, YALMIP supports SDPT3 4.0 since probably ten years back...

Cannot reproduce your issue here (running 4.0 though)

Are you really using the latest version of YALMIP?

[Alexandre Felipe]

The issue was observed when using the solver SeDuMi. In Octave 3.8.2.

[Johan Löfberg]

Yes, I used SeDuMi.

Cannot test 3.8.2, but as it works in 4.0, and in MATLAB, if you actually have the most recent version of YALMIP; I can only assume that 3.8.2 is buggy.

[Alexandre Felipe]

Do you use it under windows or Linux?
Under windows and cygwin I could run Octave + yalmip + sedumi a long time ago. But it was too slow then I gave up to try get the tings running in Python.

[Johan Löfberg]

windows