[
https://issues.apache.org/jira/browse/MATH-362?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12854056#action_12854056
]
Luc Maisonobe commented on MATH-362:
------------------------------------
Ooops. You are right.
The Levenberg-Marquardt optimizer uses specific convergence parameters which
can be set by setInitialStepBoundFactor, setCostRelativeTolerance,
setParRelativeTolerance and setOrthoTolerance.
The most important convergence tuning are either setCostRelativeTolerance for a
convergence on the cost itself or setParRelativeTolerance for a convergence on
the parameters.
I'm not sure how to solve this. Do the existing tuning parameters fit your
needs or not ? Some convergence criteria can be expressed with both methods,
but not all. Should we keep both setting as alternate methods or should we
remove one and rely on the remaining one ?
> LevenbergMarquardtOptimizer ignores the VectorialConvergenceChecker parameter
> passed to it
> ------------------------------------------------------------------------------------------
>
> Key: MATH-362
> URL: https://issues.apache.org/jira/browse/MATH-362
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 2.0
> Reporter: Roman Werpachowski
>
> LevenbergMarquardtOptimizer ignores the VectorialConvergenceChecker parameter
> passed to it. This makes it hard to specify custom stopping criteria for the
> optimizer.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
