[prev in list] [next in list] [prev in thread] [next in thread] 

List:       jakarta-commons-dev
Subject:    Re: BOBYQA Question
From:       Greg Sterijevski <gsterijevski () gmail ! com>
Date:       2011-09-30 22:35:28
Message-ID: CACMPRi0sOWKTxLxAHPcUvCr9WpoT_=W2Uw1=7K9dx7+7zN9PzQ () mail ! gmail ! com
[Download RAW message or body]


Gilles,

I have attached a test harness which shows one spot where BOBYQA fails.

See the "NonLinearConjugateGradient" thread.

Thanks,

-Greg

On Thu, Sep 29, 2011 at 6:06 PM, Gilles Sadowski <
gilles@harfang.homelinux.org> wrote:

> Hi.
>
> The work to be done on "BOBYQAOptimizer" is discussed in issue 621:
>  https://issues.apache.org/jira/browse/MATH-621
>
> There is still a lot to do on this code. And I wouldn't advise you to use
> it
> as is because the interface is likely to change: In fact, CM lacks an
> interface/base class for optimizers with constraints.
>
> >
> > I am testing some Limited Dependent Variable regressions and in shopping
> for
> > an optimizer I ran BOBYQA. I got the following exception:
> >
> > If this exception is thrown, just remove it from the code
> > org.apache.commons.math.optimization.direct.PathIsExploredException: If
> this
> > exception is thrown, just remove it from the code
> >     at
> >
> org.apache.commons.math.optimization.direct.BOBYQAOptimizer.prelim(BOBYQAOptimizer.java:1780)
> >     at
> >
> org.apache.commons.math.optimization.direct.BOBYQAOptimizer.bobyqb(BOBYQAOptimizer.java:436)
> >
> > What is this message? Is it significant?
>
> Please see the issue referred to above.
> In short, the current set of unit tests does not cover all code paths. I've
> added "throw" statements at those places which I've currently spotted
> (probably not all of them yet).
>
> So, you've run a case that exercises a branch not covered by any of the
> current tests: Hence, it would be very useful if you could make a unit test
> out of your use case, and add it to "BOBYQAOptimizerTest"!
> But please, do read the issue page: the unit tests that we have now are
> supposed to reproduce the behaviour of the original Fortran code. I think
> that any additional test (and the more the better and ideally covering all
> the code) should become part of this baseline, intended to ensure that the
> incremental code changes do not alter the efficiency (i.e. the number of
> function evalutations).
>
>
> Thanks,
> Gilles
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>


[prev in list] [next in list] [prev in thread] [next in thread]