[
https://issues.apache.org/jira/browse/NUMBERS-156?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17335484#comment-17335484
]
Alex Herbert commented on NUMBERS-156:
--------------------------------------
We should consider pros and cons of removing {{SafeNorm}}. The method only does
a single pass over the array. Also we can update the constants. The source code
states as much in [enorm.f|http://www.netlib.org/minpack/enorm.f]:
{noformat}
the definitions of small, intermediate and large components
depend on two constants, rdwarf and rgiant. the main
restrictions on these constants are that rdwarf**2 not
underflow and rgiant**2 not overflow. the constants
given here are suitable for every known computer.
{noformat}
A different implementation with power of 2 scaling has the advantage of
minimising precision loss due to divisions. But if the implementation is to
scale all numbers using the same scale factor then it requires 2 passes over
the array. So ideally we should have a test showing the precision on random
vectors and a speed test to make an informed choice.
Q. Should it be a requirement that permutations of a vector have the same
length?
> 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)
