[jira] [Commented] (MATH-993) GaussNewtonOptimizer convergence on singularity

["Gilles (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MATH-993?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13686620#comment-13686620 ] 

Gilles commented on MATH-993:
-----------------------------

Could you please provide a unit test showing the desired behaviour? Thanks.

bq. I would propose to add a convergence check into the catch-phrase, [...]

At first sight, it seems that would suffice to move the convergence check so that it happens _before_ the code block that can potentially raise the exception.

                
> GaussNewtonOptimizer convergence on singularity
> -----------------------------------------------
>
>                 Key: MATH-993
>                 URL: https://issues.apache.org/jira/browse/MATH-993
>             Project: Commons Math
>          Issue Type: Bug
>    Affects Versions: 3.2
>            Reporter: Christoph Höger
>            Priority: Minor
>
> I am (ab-)using the GaussNewtonOptimizer as a MultivariateFunctionSolver (as I could not find one in commons.math). Recently I stumbled upon an interesting behavior in one of my test cases: If a function is defined in a way that yields a minimum (a root in my case) at a singular point, the solver crashes. This is because of the following lines in doOptimize():
> catch (SingularMatrixException e) {
>                 throw new ConvergenceException(LocalizedFormats.UNABLE_TO_SOLVE_SINGULAR_PROBLEM);
>             }
> I would propose to add a convergence check into the catch-phrase, so the solver returns the solution in that special case.

--
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