[jira] [Commented] (MATH-1327) Assessment of the quality of the random number generators

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

    [ https://issues.apache.org/jira/browse/MATH-1327?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15189376#comment-15189376 ] 

Gilles commented on MATH-1327:
------------------------------

I propose that the userguide includes a new section that will report on
# performance
# "quality" assessment

of all the RNG implementations provided as part of MATH-1335.

For the first point, I'll provide micro-benchmarks produced by
* CM's {{PerfTestUtils}}
* JMH

For the second point, I'll provide reports generated by the following "stress test" suites:
* [Dieharder|http://www.phy.duke.edu/~rgb/General/dieharder.php]
* [TestU01|http://simul.iro.umontreal.ca/testu01/tu01.html]

> Assessment of the quality of the random number generators
> ---------------------------------------------------------
>
>                 Key: MATH-1327
>                 URL: https://issues.apache.org/jira/browse/MATH-1327
>             Project: Commons Math
>          Issue Type: Sub-task
>            Reporter: Gilles
>            Assignee: Gilles
>            Priority: Minor
>              Labels: correctness, documentation
>
> For some of the RNGs implemented in CM, a list of reference values (i.e. a sequence of numbers produced from using a certain seed) are provided by the algorithm's author (this is the case for "MersenneTwister").
> When available, a unit test can show that the CM implementation of the original algorithm is correct (by comparing the sequence produced by the CM code with the reference values).
> When not available, we can compile the reference implementation (usually in C), run it and convert the output (taking  care of Java "signed" vs C "unsigned" types) into a list to be used as above.
> As the intermediate steps could involve bugs, the overly cautious might wish to be offered additional assessments that the CM implementations do produce sequences of uniformly distributed numbers.



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