[
https://issues.apache.org/jira/browse/MATH-1092?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13917536#comment-13917536
]
Gilles commented on MATH-1092:
------------------------------
Parameter added (revision 1573316).
However, I'm worried that some unit tests for "PowellOptimizer" fail when this
parameter is set to a small value (1e-8). Thus, for "PowellOptimizer", I've
hard-coded the value to 1 so that the previous behaviour is retained.
Maybe there is an interpendence between this value and the other parameters of
"BracketFinder".
In your problem, are there oscillations in the vicinity of the optimum?
It is difficult for me to figure out to the appropriate improvements without a
unit test.
> NonLinearConjugateGradientOptimizer's Line search is a gradient search
> returns obviously suboptimal point.
> ----------------------------------------------------------------------------------------------------------
>
> Key: MATH-1092
> URL: https://issues.apache.org/jira/browse/MATH-1092
> Project: Commons Math
> Issue Type: Bug
> Reporter: Ajo Fod
> Attachments: MATH-1092.patch
>
>
> In package : org.apache.commons.math3.optim.nonlinear.scalar.gradient
> In a minimization problem, a line search should not return a point where the
> value is greater than the values at the edges of the interval. The line
> search violates this obvious requirement by focusing solely on solving for
> gradient=0 and ignoring the value.
> Moreover LineSearchFunction is something that can be used in other contexts,
> so perhaps this should be a standalone class.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
