[
https://issues.apache.org/jira/browse/MATH-1026?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13774547#comment-13774547
]
Gilles commented on MATH-1026:
------------------------------
Thanks.
bq. What do I need to do to move this forward?
This is a very interesting starting point, but we must think further about a
few issues:
# Enumerate here all the questions which you asked in the code (cf "// TODO"
marks).
# There seems to be a design problem:
In "OptimizationProblem", the accessor methods are not all on the same footing;
some create fresh ("Incrementor") objects, but the "ConvergenceChecker" is a
reference to an object created externally (and we discussed that those could
carry state).
Actually, I find it strange that the concepts of allowed number of
"evaluations" and "iterations" are part of the "problem" and not the
"optimizer".
# The counting is done explicitly in "LevenbergMarquardtOptimizer" whereas it
seems that the "counting option" should be enabled through a call to
"countEvaluations" in "LeastSquaresFactory".
I'd suggest that instead of an "Incrementor" argument, that method be passed
the argument currently used in "AbstractOptimizationProblem".
# "LeastSquaresProblemImpl": in CM, we avoid using the suffix "Impl".
Either it is intended that several implementations can exist (and those should
bear a name that make their specifics explicit), or there will only ever be one
implementation, and (we concluded that) it is better to use a concrete
(abstract) class rather than an interface.
# I would avoid to "import" inner classes. I think that code is easier to read
when one can readily distinguish where one should look for the definition of an
object.
# Your comment in "AbstractEvaluation" says it is designed for extension, yet
the class is package-private.
# Class "DenseWeightedEvaluation" has no associated unit tests.
> Separate Optimization Problem from Algorithm
> --------------------------------------------
>
> Key: MATH-1026
> URL: https://issues.apache.org/jira/browse/MATH-1026
> Project: Commons Math
> Issue Type: Improvement
> Reporter: Evan Ward
> Attachments: opt.tar.gz, opt.tar.gz
>
>
> See discussion on the mailing list starting with:
> http://www.mail-archive.com/[email protected]/msg39681.html
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
