Mailing-List: contact issues-help@commons.apache.org; run by ezmlm
Precedence: bulk
Reply-To: issues@commons.apache.org
Date: Fri, 25 Nov 2011 06:54:40 +0000 (UTC)
From: =?utf-8?Q?S=C3=A9bastien_Brisard_=28Commented=29_=28JIRA=29?=
 <jira@apache.org>
To: issues@commons.apache.org
Message-ID: 
 <2055717074.11044.1322204080706.JavaMail.tomcat@hel.zones.apache.org>
In-Reply-To: 
 <859386096.11914.1320783471556.JavaMail.tomcat@hel.zones.apache.org>
Subject: [jira] [Commented] (MATH-703) Splitting up the distribution
 hierarchy
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable


    [ https://issues.apache.org/jira/browse/MATH-703?page=3Dcom.atlassian.j=
ira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=3D131570=
01#comment-13157001 ]=20

S=C3=A9bastien 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 actua=
l distribution is cached, which is good, but I think that nothing guarantee=
s immutability of classes inheriting {{AbstractRealDistribution}}. For exam=
ple, I might implement my own normal distribution, with {{setMean(double)}}=
 and {{setVariance(double)}} methods, and I would then need to make sure th=
e proper private fields in the {{AbstractRealDistribution}} are updated acc=
ordingly. While caching should be allowed for in this hierarchy of classes,=
 I think no default implementation of {{getNumericalMean()}} and {{getNumer=
icalVariance()}} should be provided at this stage. This might be silly, ple=
ase 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 bas=
e abstract class.
* I'm not convinced about {{AbstractDefaultRealDistribution}} and {{Abstrac=
tDefaultIntegerDistribution}}: not having them would only mean that we woul=
d need to implement one or two additional methods, with very simple (in fac=
t, constant, as you noticed) values. So we need to compare this slight inco=
nvenience on the one hand with the encumbrance of the class hierarchy on th=
e 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.
               =20
> 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.ht=
ml), 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 re=
quire different functionality. Additionally, a super-inferface would requir=
e to parametrize the domain which makes things more complicated (e.g., "dou=
ble" would have to be replaced by "Double"). Currently, Commons Math suppor=
ts distributions with real domain and distributions with integer domain. Th=
us there will be the interfaces RealDistribution and IntegerDistribution.
> Another point is to drop the special cases of distributions with real dom=
ain in order to simplify the structure. There won't be an interface for abs=
olutely continuous distributions, and there won't be an interface for discr=
ete 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 administrato=
rs: https://issues.apache.org/jira/secure/ContactAdministrators!default.jsp=
a
For more information on JIRA, see: http://www.atlassian.com/software/jira