Re: [ojAlgo-user] using quadratic solver

[David J. <dav...@bu...>]

Fair enough.  I think the issue was just that I was trying to configure my
problem through QuadraticExpressionModel where it really wasn't necessary,
since my equation is in the standard form taken by the builder.  Going
purely by the builder's configuration methods, my code seems to be
functioning:

//now minimize (1/2*transpose(alpha)*H*alpha - sum(alpha))
        //subject to I*alpha <= C

        //build singular column vector to multiply by alpha, in order
        //to represent sum(alpha)
        double[] tempsing = new double[c];
        Arrays.fill(tempsing, 1);
        JamaMatrix sing = jfac.makeColumn(tempsing);
        //build AI = I
        JamaMatrix AI = jfac.makeEye(c, c);
        //build BI = C
        double[] BItemp = new double[c];
        Arrays.fill(BItemp, C);
        JamaMatrix BI = jfac.makeColumn(BItemp);
        //initialize alpha
        double[] alphatemp = new double[c];
        JamaMatrix alpha = jfac.makeColumn(alphatemp);

        //Set up the minimizer
        final Variable[] vars = new Variable[0];
        final QuadraticExpressionsModel qem = new
QuadraticExpressionsModel(vars);
        qem.setMinimisation(true);
        final QuadraticSolver.Builder build = new
QuadraticSolver.Builder(qem);
        build.objective(H, sing);
        build.inequalities(AI, BI);
        build.setX(alpha);
        QuadraticSolver solver = build.build();
        OptimisationSolver.Result res = solver.solve();

On Wed, Oct 6, 2010 at 4:05 PM, Anders Peterson <an...@op...>wrote:

> The code you've written declares an optimisation problem with 2 variables,
> but the variable H is not part of the objective function, not part of any
> expression and not constrained in any way.
>
> Actually I think you ask too many things at the same time; how can you not
> know if alpha and H should be "listed" as variables or not? Are they
> variables?
>
> You make something (a single variable, or a whole expression) part of the
> objective function by setting its contribution weight. What you set
> individual weights to only matters in relation other weights.
>
> You constrain something (single variable, or whole expressions) by setting
> lower and/or upper limits.
>
> /Anders
>
>
> On 6 okt 2010, at 01.17, David Johnson wrote:
>
> > Hi, I'm new to ojalgo, and trying to work out the exact syntax needed to
> configure the quadratic solver.
> >
> > Specifically, I'm trying to solve a standard quadratic minimization
> problem: min(1/2*transpose(alpha)*H*alpha - sum(alpha)), given 0 <= alphai
> <= C (where C is a constant), for each element in alpha.  Alpha will be
> initialized to 0s (or uninitialized, if no initial input is needed) and H to
> a given input matrix.  I'm unclear on exactly what I need to configure in
> the QuadraticExpressionsModel versus what is handled by the builder.  I also
> can't seem to find a current example of how to configure an expression-based
> model (except for the one in MarkowitzModel, which helped, but didn't
> entirely clarify the logic).
> >
> > My current (very rough) code skeleton looks like this:
> >
> > final Variable[] vars = new Variable[2];
> >         vars[0] = new Variable("alpha");
> >         vars[0].setLowerLimit(new BigDecimal(0));
> >         vars[0].setUpperLimit(new BigDecimal(C));
> >         vars[0].setContributionWeight(BigMath.ONE);
> >         vars[1] = new Variable("H");
> >
> >         final QuadraticExpressionsModel qem = new
> QuadraticExpressionsModel(vars);
> >         qem.setMinimisation(true);
> >
> >         final QuadraticSolver.Builder build = new
> QuadraticSolver.Builder(qem);
> >
> > I can't tell if alpha and H should even be listed as variables (as
> opposed to expressions), how to associate my input matrices with them if
> they should be, and how to describe range limits on alpha if they should
> not.  I'm also not certain how to define my objective function - is the
> basic format handled by the builder, or do I need to describe it in the
> model somehow?  Should I be using the inequalities(...) and objective(...)
> functions in the builder, or defining those via the model?
> >
> > Also, I know I can represent sum(alpha) in my equation by just setting C
> (the vector in the basic equation format, not my constant) to be a singular
> 1 vector, but is there any way to ensure that the solver will then treat C
> as a constant?
> >
> > Thanks
> > David Johnson
> >
> ------------------------------------------------------------------------------
> > Beautiful is writing same markup. Internet Explorer 9 supports
> > standards for HTML5, CSS3, SVG 1.1,  ECMAScript5, and DOM L2 & L3.
> > Spend less time writing and  rewriting code and more time creating great
> > experiences on the web. Be a part of the beta today.
> >
> http://p.sf.net/sfu/beautyoftheweb_______________________________________________
> > ojAlgo-user mailing list
> > ojA...@li...
> > https://lists.sourceforge.net/lists/listinfo/ojalgo-user
>
>
>
> ------------------------------------------------------------------------------
> Beautiful is writing same markup. Internet Explorer 9 supports
> standards for HTML5, CSS3, SVG 1.1,  ECMAScript5, and DOM L2 & L3.
> Spend less time writing and  rewriting code and more time creating great
> experiences on the web. Be a part of the beta today.
> http://p.sf.net/sfu/beautyoftheweb
> _______________________________________________
> ojAlgo-user mailing list
> ojA...@li...
> https://lists.sourceforge.net/lists/listinfo/ojalgo-user
>