[
https://issues.apache.org/jira/browse/NUMBERS-156?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17343887#comment-17343887
]
Alex Herbert commented on NUMBERS-156:
--------------------------------------
Although the mean error is lower for enormMod than enorm the standard deviation
of enormMod is worse than enorm when it always scales (low or high). It matches
that of the mid dataset which is what we would expect as scaling is exact.
However the mean is different which I do not understand. If scaling is exact
then the ulp error should be the same.
Can you try the modified version with Kahan summation? I am interested in
whether it is worth it to try and reduce the ULP min/max error. I do not think
it will be much slower as the number of conditional statements is the same.
> SafeNorm 3D overload
> --------------------
>
> Key: NUMBERS-156
> URL: https://issues.apache.org/jira/browse/NUMBERS-156
> Project: Commons Numbers
> Issue Type: Improvement
> Reporter: Matt Juntunen
> Priority: Major
>
> We should create an overload of {{SafeNorm.value}} that accepts 3 arguments
> to potentially improve performance for 3D vectors.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
