[
https://issues.apache.org/jira/browse/MATH-1137?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14062190#comment-14062190
]
Gilles commented on MATH-1137:
------------------------------
bq. I am less concerned about the line we have been discussing here, than I am
about other bugs which we are not discussing and have not yet been found!
bq. You can often have optimisation algorithms which have bugs but still find
the minimum, albeit inefficiently.
A general point worth raising on the "dev" ML!
bq. \[...\] was there a comparison between Java and the Fortran to see if the
code reproduced the same results at each stage, not just the final result?
Dietmar could probably answer. He might still read the "dev" ML, so better to
ask there.
IIRC, the tolerances in the unit/validation tests are so stringent that even
seemingly innocuous changes made some tests fail; hence it is unlikely that
different paths would not trigger a failure. Although this might reassure you,
it is a liability in the way to the refactoring (see MATH-621).
> BOBYQA incorrect indexing
> -------------------------
>
> Key: MATH-1137
> URL: https://issues.apache.org/jira/browse/MATH-1137
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.3
> Reporter: Nigel Goodwin
>
--
This message was sent by Atlassian JIRA
(v6.2#6252)
