You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@commons.apache.org by "Gilles (JIRA)" <ji...@apache.org> on 2016/03/20 19:05:33 UTC
[jira] [Updated] (MATH-1344) "ZipfDistribution" cleanup
[ https://issues.apache.org/jira/browse/MATH-1344?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Gilles updated MATH-1344:
-------------------------
Issue Type: Sub-task (was: Improvement)
Parent: MATH-1345
> "ZipfDistribution" cleanup
> --------------------------
>
> Key: MATH-1344
> URL: https://issues.apache.org/jira/browse/MATH-1344
> Project: Commons Math
> Issue Type: Sub-task
> Reporter: Gilles
> Priority: Minor
> Labels: cleanup, performance, private
> Fix For: 4.0
>
>
> * We should avoid hard-coded numbers.
> * Inner class {{ZipfRejectionInversionSampler}} should be private.
> It is in package scope only for the sake of testing "helper" functions that are not specific to that implementation.
> * The helper functions are
> {noformat}
> log1p(x) / x
> {noformat}
> and
> {noformat}
> expm1(x) / x
> {noformat}
> but with a special code path if x < 1e-8, where the Taylor expansion
> is used.
> If this leads to more correct results (or is more efficient), it should be used generally by moving the optimization to the corresponding "function object" defined in {{o.a.c.m.analysis.function}} (as is done for e.g. the {{Sinc}} function).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)