N-Queens problem with Yalmip

[Asher Metzger]

Hey Johan,

I've been trying to solve the 8-queen (placing 8 queens on a chess board without having them threaten each other)  placement problem with Yalmip.
I run in to 2 problems:

What should be my objective function?

and why are the constraints trivally false?

thanks,

Asher (attaching the code I tried)

[Johan Löfberg]

row(1) ~= row(1) sounds like a generally bad idea.

Well, if you don't know what you want to have as objective, why do you want an objective?

[Asher Metzger]

Of course.

I didn't upload the updated file. 

now row (i) and row (j) don't have the same index. yet the answer I get is a vector of four's which is against the first constraint. 

[Johan Löfberg]

The logical code is in a state of constant development (i.e. buggy)

Change line 30 in @sdpvar/ne to

elseif isa(X,'sdpvar') & is(X,'integer')

and it might work

[Asher Metzger]

Now it runs. But it seems to run forever.

I put a 5000 iteration limit but it still goes.

[Johan Löfberg]

Which solver? Solved in 0s with all solvers here

[Asher Metzger]

without C2 or C3 it finds a solution by 100 iter's but not with both

[Johan Löfberg]

Which solver?

[Asher Metzger]

With BNB.

Attached is the file with the corrected indicing.

[Johan Löfberg]

Don't use bnb for milps
http://users.isy.liu.se/johanl/yalmip/pmwiki.php?n=Solvers.BNB

[Asher Metzger]

I tried with mosek, Works like a charm. 

Thank you

[Mark L. Stone]

With the change to @sdpvar/n, it seemed to run correctly with several solvers using YALMIP R20141218 (with not surprisingly, different valid solutions depending on the solver), but produces a vector of ones using R20150204.  But as you say, perhaps that code is still buggy.

But my bigger discovery was an apparent bug in the interface to baron when there is no objective function or a fake objective function such as 0 which has no sdpvar.
Here is the simplest example of that bug:

sdpvar x;optimize(x>=0,[],sdpsettings('solver','baron','debug',1))

which produces

Error using callbaron (line 23)
Error: Expression or statement is incomplete or incorrect.
Error in solvesdp (line 329)
    eval(['output = ' solver.call '(interfacedata);']);
Error in optimize (line 31)
varargout{:} = solvesdp(varargin{:});

Use of 0 instead of [] for the objective gives the same error.

sdpvar x;optimize(x>=0,x,sdpsettings('solver','baron','debug',1))
executes correctly

I am using matbar version 1.69.

[Johan Löfberg]

Thank you. Change line 22 in callbaron.m to

if length(obj)>0
    obj = ['@(x) ' obj];
    obj = eval(obj);
else
    obj = @(x)0;
end

[Mark L. Stone]

Thanks.  I presume that is the replacement collectively for lines 22-23.

BTW, ignore my previous comment about getting the wrong answer with R20150204. I had a late night screw up with my file naming, so got the wrong result when using what turned out to be the uncorrected version of @sdpvar/ne.