[jira] Commented: (MATH-521) Changes in "HarmonicCoefficientsGuesser"

["Luc Maisonobe (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MATH-521?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12997696#comment-12997696 ] 

Luc Maisonobe commented on MATH-521:
------------------------------------

Do you want to remove the guess method or to have default values just when the guess fails ? We should add tests for such failures. I think one way to make it fail would be for example to supply a constant sample with 0 values only for y, or a set of points with only one value for x and random values for y. 

I think for the frequency part, 0 is probably not a good choice and would lead to numerical problems. Perhaps a value like 2PI/l where l is the range in X would be better, it would start considering the sample covers exactly one period.

The javadoc is for sure not perfect but is readable in a browser, which is the main intend. The output is here: [http://commons.apache.org/math/api-2.1/org/apache/commons/math/optimization/fitting/HarmonicCoefficientsGuesser.html].

Putting this in the user guide would be a nice addition.

> Changes in "HarmonicCoefficientsGuesser"
> ----------------------------------------
>
>                 Key: MATH-521
>                 URL: https://issues.apache.org/jira/browse/MATH-521
>             Project: Commons Math
>          Issue Type: Improvement
>            Reporter: Gilles
>            Assignee: Gilles
>            Priority: Minor
>              Labels: api-change, documentation
>             Fix For: 3.0
>
>
> (1) The "guess" method throws "OptimizationException" when the algorithm fails to determine valid values for amplitude and angular frequency.
> There are no test showing how such a situation can occur.
> Moreover, since this procedure is used to provide an initial guess to an optimizer, it is better to pick any values for those parameters (e.g. zero) and let the optimizer proceed from that initial point.
> (2) The class javadoc seems very thorough in explaining the algorithm, but is quite unreadable in the source code, making it fairly useless for checking how the code complies with the comments. I think that this explanation should go in the user guide (and leave a mostly "plain text" outline of the algorithm, referring to the guide for details). [Does the format of the user guide allow such tricky (ASCII "art") constructs?]

-- 
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira