You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@commons.apache.org by Phil Steitz <ph...@gmail.com> on 2015/01/09 18:55:53 UTC
[math] tail probabilites
There is a feature request implicit in MATH-1185 - to add direct
computation of (upper) tail probabilities, so users can avoid loss
of significance errors when computing them from cumulative
probabilities, which is the only thing they can do with current
implementations. In some cases, we can fairly easily provide direct
estimates; in others, not so easy.
What might make sense would be:
* Add an upperTail(x) method to the 4.0 distributions interface
(maybe with a better name)
* Add default implementation to AbstractXxDistribution (3.x and 4)
that returns naive results
* Add - and doc - better implementations for individual
distributions as and when we get them
Phil
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [math] tail probabilites
Posted by Phil Steitz <ph...@gmail.com>.
On 1/11/15 10:04 AM, Phil Steitz wrote:
> On 1/11/15 9:26 AM, Gilles wrote:
>> On Fri, 09 Jan 2015 10:55:53 -0700, Phil Steitz wrote:
>>> There is a feature request implicit in MATH-1185 - to add direct
>>> computation of (upper) tail probabilities, so users can avoid loss
>>> of significance errors when computing them from cumulative
>>> probabilities, which is the only thing they can do with current
>>> implementations. In some cases, we can fairly easily provide direct
>>> estimates; in others, not so easy.
>>>
>>> What might make sense would be:
>>>
>>> * Add an upperTail(x) method to the 4.0 distributions interface
>>> (maybe with a better name)
>>> * Add default implementation to AbstractXxDistribution (3.x and 4)
>>> that returns naive results
>>> * Add - and doc - better implementations for individual
>>> distributions as and when we get them
>> +1
>> [Although we could consider throwing an exception instead of
>> returning the naive result. Or one method that would throw
>> and one that would return possibly inaccurate results.]
> Yeah, UnsupportedOperationException might make sense in place of the
> naive implementation. Would have to be documented carefully if we
> decide to do that.
I guess another idea would be not to add the method to the
interfaces, but just to the distributions that have good
implementations.
Phil
>
> Phil
>>
>> Gilles
>>
>>> Phil
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
>> For additional commands, e-mail: dev-help@commons.apache.org
>>
>>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [math] tail probabilites
Posted by Phil Steitz <ph...@gmail.com>.
On 1/11/15 9:26 AM, Gilles wrote:
> On Fri, 09 Jan 2015 10:55:53 -0700, Phil Steitz wrote:
>> There is a feature request implicit in MATH-1185 - to add direct
>> computation of (upper) tail probabilities, so users can avoid loss
>> of significance errors when computing them from cumulative
>> probabilities, which is the only thing they can do with current
>> implementations. In some cases, we can fairly easily provide direct
>> estimates; in others, not so easy.
>>
>> What might make sense would be:
>>
>> * Add an upperTail(x) method to the 4.0 distributions interface
>> (maybe with a better name)
>> * Add default implementation to AbstractXxDistribution (3.x and 4)
>> that returns naive results
>> * Add - and doc - better implementations for individual
>> distributions as and when we get them
>
> +1
> [Although we could consider throwing an exception instead of
> returning the naive result. Or one method that would throw
> and one that would return possibly inaccurate results.]
Yeah, UnsupportedOperationException might make sense in place of the
naive implementation. Would have to be documented carefully if we
decide to do that.
Phil
>
>
> Gilles
>
>> Phil
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org
Re: [math] tail probabilites
Posted by Gilles <gi...@harfang.homelinux.org>.
On Fri, 09 Jan 2015 10:55:53 -0700, Phil Steitz wrote:
> There is a feature request implicit in MATH-1185 - to add direct
> computation of (upper) tail probabilities, so users can avoid loss
> of significance errors when computing them from cumulative
> probabilities, which is the only thing they can do with current
> implementations. In some cases, we can fairly easily provide direct
> estimates; in others, not so easy.
>
> What might make sense would be:
>
> * Add an upperTail(x) method to the 4.0 distributions interface
> (maybe with a better name)
> * Add default implementation to AbstractXxDistribution (3.x and 4)
> that returns naive results
> * Add - and doc - better implementations for individual
> distributions as and when we get them
+1
[Although we could consider throwing an exception instead of
returning the naive result. Or one method that would throw
and one that would return possibly inaccurate results.]
Gilles
> Phil
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org