[
https://issues.apache.org/jira/browse/MATH-1351?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15214968#comment-15214968
]
Gilles commented on MATH-1351:
------------------------------
Code available in branch "feature-MATH-1158".
One now failing test has been "Ignore"d due to the (IMO invalid) expectation
that the sequence of "double" being used will always be the same across CM
evolutions.
In this unit test case, the failure is due to the transformation of the seed
value (which has changed in the new {{o.a.c.m.rng}} package - cf. MATH-1335),
leading to a different sequence of numbers.
If the unit test is intended to ensure that some sequence is not changed (e.g.
by the introduction of a bug in the code of the class under test), then I'd
assume that letting the current code generate a list of "reference" values, and
compare against that, is enough.
However what the test intended to ensure is that a set of samples is "correct"
if it is compatible with the distribution's parameters (this check being done
with another software). This latter goal could perhaps be achieved with the
same level of confidence (cf. accuracy of the retrieved parameters) using only
CM.
> New sampler API for "MultivariateRealDistribution"
> --------------------------------------------------
>
> Key: MATH-1351
> URL: https://issues.apache.org/jira/browse/MATH-1351
> Project: Commons Math
> Issue Type: Sub-task
> Reporter: Gilles
> Assignee: Gilles
> Priority: Minor
> Labels: api, deprecation
> Fix For: 4.0
>
>
> Make changes similar to those {{RealDistribution}} and
> {{IntegerDistribution}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
