[
https://issues.apache.org/jira/browse/MATH-1024?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13748599#comment-13748599
]
Gilles commented on MATH-1024:
------------------------------
I actually meant a test case involving the optimizer, that leads to a seemingly
inconsistent behaviour from the user viewpoint.
What you advocate here, IIUC, is that the default threshold be set to the same
value in the decomposition classes. This is not a bug per se since the
documentation mentions it.
As I indicate above, maybe that it leads to a bug in "GaussNewtonOptimizer", in
which case it should be corrected there (by setting an explicit threshold).
Please open a new issue for that.
As for this request, you should raise it on the "dev" ML.
If people there agree to the change of default, it probably cannot be
implemented before version 4.0.
> LU and QR have different default singularity thresholds
> -------------------------------------------------------
>
> Key: MATH-1024
> URL: https://issues.apache.org/jira/browse/MATH-1024
> Project: Commons Math
> Issue Type: Bug
> Affects Versions: 3.2
> Reporter: Evan Ward
> Priority: Minor
> Attachments: ToleranceTest.java
>
>
> LU uses 1e-11 and QR uses 0. This means by default QR never throws a
> Singularity exception. This caused divergence to go unreported in
> GaussNewtonOptimizer when it uses QR. I think a consistent default should be
> chosen. I will make the threshold explicit in GaussNewton.
--
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
