[
https://issues.apache.org/jira/browse/MATH-385?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12888583#action_12888583
]
Mikkel Meyer Andersen commented on MATH-385:
--------------------------------------------
I agree that we need to handle non-connected support, and a isSupportConnected
might be a way to do it (dealing with it properly in {Lower,
Upper}Bound-functions if it is not connected). Or else we should think of
support as an array of supports, but I think that would complicate it all too
much. But the idea might give rise to an Interval-class consisting of endpoints
(lower and upper bounds) together with indication of whether they are inclusive
or not, and whether the interval is continuous or discrete.
I would never expect to see NaN in support (I'm not sure if my expectations are
valid), but in mean and variance it is very reasonable to expect a NaN, e.g. is
neither mean nor variance defined for the Cauchy distribution.
Regarding the significance of the computations, you might be right. But I think
it should be possible to specify not to do it before get()'ting the mean etc.
E.g. is the mean for the Weibull constructed by computing a sum which might
contain many terms depending on the parameters. It would not be a problem in
3.0 when removing the ability to change the parameters. If we simply add a
boolean isMeanCalculated etc. this behavior can easily be obtained.
> Characteristic (support, mean, variance, ...) on Distributions
> --------------------------------------------------------------
>
> Key: MATH-385
> URL: https://issues.apache.org/jira/browse/MATH-385
> Project: Commons Math
> Issue Type: New Feature
> Reporter: Mikkel Meyer Andersen
> Original Estimate: 5h
> Remaining Estimate: 5h
>
> I wish that the Distributions could contain some characteristics. For example
> support, mean, and variance.
> Support:
> AbstractContinuousDistribution and AbstractIntegerDistribution should have
> double getSupport{Lower, Upper}Bound() and int getSupport{Lower,
> Upper}Bound(), respectively. Also methods a la boolean isSupport{Lower,
> Upper}BoundInclusive() on AbstractContinuousDistribution should reflect if
> the support is open of closed. In practise the implemented distributions are
> easy since the support for all continuous distributions are real intervals
> (connected sets), and the support for all the discrete distributions are
> connected integer sets. This means that the lower and upper bound (together
> with isSupport{Lower, Upper}BoundInclusive() on
> AbstractContinuousDistribution because it is not needed on the discrete
> distributions because of their nature) are sufficient for determine the
> support.
> Mean and variance:
> double get{Mean, Variance}() should be on AbstractDistribution.
> With such characteristic an invalidateParameters-method might come in handy
> because they often depend on the parameters. The characteristics should not
> be calculated before the first time they are get'ted, and when calculated,
> they should be saved for later use. When parameters change, an
> invalidateParameters-method should be called to force the characteristics to
> be recalculated.
> Values such as Double.POSITIVE_INFINITY, Double.NEGATIVE_INFINITY, and
> Double.NaN should be used where appropriate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
