[jira] [Commented] (MATH-1346) "ValueServer" has various shortcomings

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

    [ https://issues.apache.org/jira/browse/MATH-1346?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15287198#comment-15287198 ] 

Gilles commented on MATH-1346:
------------------------------

Class is going to be removed (cf. [this message to the ML|http://markmail.org/message/7t2uvrqldgynm2ds]).

> "ValueServer" has various shortcomings
> --------------------------------------
>
>                 Key: MATH-1346
>                 URL: https://issues.apache.org/jira/browse/MATH-1346
>             Project: Commons Math
>          Issue Type: Sub-task
>            Reporter: Gilles
>            Priority: Minor
>              Labels: api, deprecation, outdated
>             Fix For: 4.0
>
>
> * It is inefficient as it recreates a distribution instance for each sample (through calls to "RandomDataGenerator").
> * It seems to be an aggregation of "EmpiricalDistribution, "ExponentialDistribution", "NormalDistribution" and "UniformDistribution".
> * Main purpose seems to allow sampling from those distributions where the parameters ("mu" and "sigma") are mutable (but are set to identical values for all distributions).
> * Then, there is a basic problem in that those instance variables are not always meaningful (for the different modes): field "sigma" only applies to the "Gaussian mode" and "mu" does not apply to the "digest mode".
> * The only non-trivial addition (wrt using the distribution classes directly) is the file IO but  usability is reduced because the class imposes a specific format and encoding.
> * The name is deceiving as one could easily assume that the server can be accessed from multiple "client" threads.
> IMO, a math library should not be concerned with how data is loaded from external sources (such as files), hence this class should perhaps be moved to the "examples" part of the repository.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)