commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Phil Steitz <phil.ste...@gmail.com>
Subject Re: [math] Package location for RankingAlgorithm
Date Sun, 12 Apr 2009 17:55:15 GMT
Luc Maisonobe wrote:
> Phil Steitz a écrit :
>   
>> MATH-136 introduces rank correlation, with pluggable ranking algorithm. 
>> The RankingAlgorithm interface and associated implementations are likely
>> to be reused elsewhere in the stat package.  The question is where to
>> put RankingAlgorithm and its implementations.  I would appreciate
>> feedback on the following alternatives.
>>
>> 0) Hold off introducing the algorithms at all - just hard-code the
>> conventional (ties get the average) algorithm into the Spearman's
>> correlation class to be included in the correlation package.  See
>> comments in the JIRA issue.
>>
>> 1) Put RankingAlgorithm and its implementations into the correlation
>> package, where it will be first used.
>>
>> 2) Put them in util
>>
>> 3) Put them in stat.descriptive.rank
>>
>> 4) new package stat.ranking
>>
>> I think 2) is the best,  but appreciate feedback.
>>     
>
> I would have chosen 4, but my lack of understanding about anything stat
> related implies my opinion should not count here. The rationale is
> probably where users without a priori knowledge of the library layout
> would first search for such features. Are ranking algorithms used
> outside of stat ?
>   
None that we currently have, though rank transformation is a pretty 
basic utility.  Within stat, there are lots of non-parametric techniques 
(such as Spearman's rank correlation) that apply rank transformations.   
FWIW, R puts their version of this in the "base" package, which is sort 
of like util for us.  I am fine with 4) and would even be OK with 0), 
since the current case does not really require it (i.e., I don't know 
how meaningful the statistic is if you don't use ties-averaged as the 
ranking algorithm).  It will be useful, however, when we implement other 
nonparametric stats to have these algorithms available.

Phil
>
>   
>> Phil
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
>> For additional commands, e-mail: dev-help@commons.apache.org
>>
>>
>>     
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>   


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Mime
View raw message