Re: [boost] math statistical distribution: multivariate gaussian

Subject: Re: [boost] math statistical distribution: multivariate gaussian
From: John Maddock (john_at_[hidden])
Date: 2008-11-19 11:09:49

Thijs van den Berg wrote:
>>> There's nothing in development at present (that I know of), so a
>>> submission would certainly be welcome.
>>> The main open issue IMO is getting the interface right for the
>>> mutivariate case.
>>>
>> Yes, exactly, the interface is one of the main reason for me to team
>> up
>> with C++ pears, to try and get that right. That a thing where
>> feedback
>> is very helpful!

Nod.

>> As I see it, there are a couple of interface issue:
>> * containers for the multivariate parameters (vectors and matrices)
>> and operators on those containers -inverse of a matrix-

Can we not define a *Concept* for multivariate parameters, and convert them 
internally to whatever internal representation is used?

Boost.uBlas has matrix inversion (via LU decomposition), but it's not 
necessarily all that stable :-(  Is matrix inversion really necessary?

>> * specifying generic functions/properties of multivariate densities,
>> that would apply to *any* multivariate density

Nod.  The list that applies to the univariate densities in Boost.Math would 
be a good place to start, modulo the ones that cannot be implemented in the 
multivariate case (quantile for example).

>> * the number of Gaussian identities/operators/function/algorithms to
>> include, there are many, some widely used, some obscure...
>>
>> I'm trying to form a team -asking friends- and will probably be able
>> to
>> get some knowledgeable C++ developers to contribute to this. As I
>> understand, the procedure is to get it working first & then submit
>> it to
>> the sandbox?

Yep.

>> How would I be able to get boost developers involved
>> during
>> the early interface design phase?

Ask here for comments on the design right here: it helps a great deal to 
have some docs at this point, as it's easier for folks to get their heads 
around.  You're also most welcome to fly your ideas past myself, Paul 
(Bristow), and whoever else volunteers first of course :-)

Cheers, John.