commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From S├ębastien Brisard (Commented) (JIRA) <j...@apache.org>
Subject [jira] [Commented] (MATH-703) Splitting up the distribution hierarchy
Date Fri, 25 Nov 2011 06:54:40 GMT

    [ https://issues.apache.org/jira/browse/MATH-703?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13157001#comment-13157001
] 

S├ębastien Brisard commented on MATH-703:
----------------------------------------

Hi Christian,
I've reviewed your patch, and have some concerns
* in class {{AbstractRealDistribution}}, the mean and variance of the actual distribution
is cached, which is good, but I think that nothing guarantees immutability of classes inheriting
{{AbstractRealDistribution}}. For example, I might implement my own normal distribution, with
{{setMean(double)}} and {{setVariance(double)}} methods, and I would then need to make sure
the proper private fields in the {{AbstractRealDistribution}} are updated accordingly. While
caching should be allowed for in this hierarchy of classes, I think no default implementation
of {{getNumericalMean()}} and {{getNumericalVariance()}} should be provided at this stage.
This might be silly, please correct me if I'm wrong. I think people raised the same concerns
about the dimensions of the domain and codomain of linear operators, when I first submitted
an implementation where these dimensions where stored in the base abstract class.
* I'm not convinced about {{AbstractDefaultRealDistribution}} and {{AbstractDefaultIntegerDistribution}}:
not having them would only mean that we would need to implement one or two additional methods,
with very simple (in fact, constant, as you noticed) values. So we need to compare this slight
inconvenience on the one hand with the encumbrance of the class hierarchy on the other hand...
And in any case, people would *always* have to refer to the Javadoc of these default implementations
to make sure that what they want to implement is indeed a {{AbstractDefaultXxxDistribution}}

These are only my views, though, and I'm going to post an invitation on the mailing-list in
order to invite others to review your work.
                
> Splitting up the distribution hierarchy
> ---------------------------------------
>
>                 Key: MATH-703
>                 URL: https://issues.apache.org/jira/browse/MATH-703
>             Project: Commons Math
>          Issue Type: Improvement
>            Reporter: Christian Winter
>            Priority: Minor
>         Attachments: MATH-703_patch.zip
>
>
> As discussed on the mailing list (http://apache-commons.680414.n4.nabble.com/math-Distributions-over-sample-spaces-other-than-R-tp3931349p3931349.html),
the distribution interfaces should be restructured.
> The most important point is to create one root interface for each domain. There should
*not* be a common super-interace because different domains require different functionality.
Additionally, a super-inferface would require to parametrize the domain which makes things
more complicated (e.g., "double" would have to be replaced by "Double"). Currently, Commons
Math supports distributions with real domain and distributions with integer domain. Thus there
will be the interfaces RealDistribution and IntegerDistribution.
> Another point is to drop the special cases of distributions with real domain in order
to simplify the structure. There won't be an interface for absolutely continuous distributions,
and there won't be an interface for discrete distributions on the real domain. All the functionality
required by the special cases can be defined in RealDistribution.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

       

Mime
View raw message