[
https://issues.apache.org/jira/browse/MATH-441?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12931829#action_12931829
]
Luc Maisonobe commented on MATH-441:
------------------------------------
The MultiStart case was put this way because some start points may be out of
bonds whereas other start points may lead to a feasible result.
So when an exception occurs for run number i, we simply set optima[i] to null
and try another point. Once the total number of runs have been performed, we
sort the array ensuring the null points are at the end and the feasible point
at the beginning (see the private sortPairs method).
In fact, this is definitely a feature of multi-start since it allows to get
around some difficult problems for which convergence is not guaranteed. It is
documented in the javadoc (perhaps this documentation should be clarified,
though).
One way to improve things and not swallow a programming error while still
circumventing unstable functions would be to rethrow the last caught exception
if it appears none of the starting points did provide a solution. So if only
some start points fail and other lead to a result, we provide the result and
ignore some exceptions, as documented, but if all points fail then there is a
real problem and we forward it above.
> Remove occurences of "FunctionEvaluationException"
> --------------------------------------------------
>
> Key: MATH-441
> URL: https://issues.apache.org/jira/browse/MATH-441
> Project: Commons Math
> Issue Type: Task
> Reporter: Gilles
> Assignee: Gilles
> Priority: Minor
> Fix For: 3.0
>
>
> As per the issue MATH-440, {{FunctionEvaluationException}} is now redundant
> and is going to removed.
> We should remove the "throws" clauses that refer to it from all the methods'
> signature.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
