[
https://issues.apache.org/jira/browse/NUMBERS-156?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17340517#comment-17340517
]
Matt Juntunen commented on NUMBERS-156:
---------------------------------------
bq. I assume the ulp error for the unsafe method includes the cases where the
result was not finite.
Yes, I thought I had corrected that but I guess not. Here is a better run.
||name||ulp error||std dev||non-finite||
|exact|0.00|0.00|0
|unsafe|0.000313|1.05e-06|10534|
|safe|0.0132|4.18e-05|0|
|herbert|0.000130|4.11e-07|0|
bq. I find it strange the unsafe method is slower.
Huh? Direct computation ignoring overflow/underflow was the fastest in the
benchmark.
bq. Are you suggesting something like ...
Yes, having a {{Norms}} class like that is what I was picturing. We don't
necessarily need all of those methods, though. I currently only have a use case
for the 3D euclidean norm and I think we should retain the array-based
euclidean norm from {{SafeNorm}}. Besides those two methods, I thought it might
be good to add other standard norms, even if they are trivial.
bq. Is there a use case for a safe implementation of the p-norm?
Not from me. I didn't know it existed until just now :-)
bq. Shouldn't we rather opt for functional interfaces:
Not sure what you mean. I don't want to have to instantiate an object to
perform these computations. Also, the static methods could be used in a
functional context as method references.
> 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)
