mutlivariate normal and t distributions

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

mutlivariate normal and t distributions

Adelchi Azzalini

Dear R-help list members,

I have created a package 'mnormt' with facilities for the multivariate
normal and t distributions. The core part is simply an interface to
Fortran routines by Alan Genz for computing the integral of two
densities over rectangular regions, using an adaptive integration
method. Other R functions compute densities and generate random
numbers.

The starting motivation to write it was the  need to have functions
which compute the distribution functions in a non-Monte Carlo form,
sinse this caused me problems when these probabilities are involved in
a minimization problem. For this reason, I could not make use of the
CRAN package 'mvtnorm'.  Exactly to avoid superposition with the CRAN
package, 'mnormt' is made available somehere else, in case other
people want to use it.  The package 'mnormt' is at
   http://azzalini.stat.unipd.it/SN/Pkg-mnormt

As explained, this is not uploaded to CRAN just to avoid clash with
the existing package. However, if it is felt appropriate, I have no
objection to upload it to CRAN.

Best wishes,

Adelchi Azzalini
--
Adelchi Azzalini  <[hidden email]>
Dipart.Scienze Statistiche, Università di Padova, Italia
tel. +39 049 8274147,  http://azzalini.stat.unipd.it/

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Reply | Threaded
Open this post in threaded view
|

Re: mutlivariate normal and t distributions

Spencer Graves
          Thanks for this.  I got stuck some months ago, because I needed to
evaluate a certain special normal tail region in at least 10,000
dimensional space, and "mvtnorm" could not give me adequate numerical
precision above about 30 or 35 dimensions.  When I get time to return to
that project, I will be particularly interested in looking at your code.

          I would strongly encourage you to submit your package to CRAN.
Otherwise it will be harder for people to learn about it and get it.  If
your package contains functions or other objects with the same names as
objects in other packages, that fact could create a conflict, though not
insurmountable.  If you can make some effort to minimize the potential
for name conflicts with other packages, that would be great.  However,
even without that, I prefer accessibility.

          Of course, from a user's perspective, it would be best if you and the
'mvtnorm' maintainer (Torsten Hothorn) could develop a consensus on the
circumstances under which each algorithm is best and then develop a
combined package that makes it easier for users to access the best known
method -- and to easily switch between methods, when it may not be clear
which is best.  However, if you and Torsten don't have time & energy for
that, I would still encourage you to submit your package to CRAN.

          Best Wishes,
          Spencer Graves

Adelchi Azzalini wrote:

> Dear R-help list members,
>
> I have created a package 'mnormt' with facilities for the multivariate
> normal and t distributions. The core part is simply an interface to
> Fortran routines by Alan Genz for computing the integral of two
> densities over rectangular regions, using an adaptive integration
> method. Other R functions compute densities and generate random
> numbers.
>
> The starting motivation to write it was the  need to have functions
> which compute the distribution functions in a non-Monte Carlo form,
> sinse this caused me problems when these probabilities are involved in
> a minimization problem. For this reason, I could not make use of the
> CRAN package 'mvtnorm'.  Exactly to avoid superposition with the CRAN
> package, 'mnormt' is made available somehere else, in case other
> people want to use it.  The package 'mnormt' is at
>    http://azzalini.stat.unipd.it/SN/Pkg-mnormt
>
> As explained, this is not uploaded to CRAN just to avoid clash with
> the existing package. However, if it is felt appropriate, I have no
> objection to upload it to CRAN.
>
> Best wishes,
>
> Adelchi Azzalini

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Reply | Threaded
Open this post in threaded view
|

Re: mutlivariate normal and t distributions

torsten-5

On Mon, 23 Jan 2006, Spencer Graves wrote:

>  Thanks for this.  I got stuck some months ago, because I needed to
> evaluate a certain special normal tail region in at least 10,000
> dimensional space, and "mvtnorm" could not give me adequate numerical
> precision above about 30 or 35 dimensions.  When I get time to return to
> that project, I will be particularly interested in looking at your code.
>
>  I would strongly encourage you to submit your package to CRAN.

yes, Adelchi, you should definitely submit the package to CRAN!

> Otherwise it will be harder for people to learn about it and get it.  If
> your package contains functions or other objects with the same names as
> objects in other packages, that fact could create a conflict, though not
> insurmountable.  If you can make some effort to minimize the potential
> for name conflicts with other packages, that would be great.  However,
> even without that, I prefer accessibility.
>
>  Of course, from a user's perspective, it would be best if you and the
> 'mvtnorm' maintainer (Torsten Hothorn) could develop a consensus on the
> circumstances under which each algorithm is best and then develop a
> combined package that makes it easier for users to access the best known
> method -- and to easily switch between methods, when it may not be clear
> which is best.  However, if you and Torsten don't have time & energy for
> that, I would still encourage you to submit your package to CRAN.
>

we already talked about such ideas but, for the time being, having two
packages should not be a big problem.

Best,

Torsten


>  Best Wishes,
>  Spencer Graves
>
> Adelchi Azzalini wrote:
>
> > Dear R-help list members,
> >
> > I have created a package 'mnormt' with facilities for the multivariate
> > normal and t distributions. The core part is simply an interface to
> > Fortran routines by Alan Genz for computing the integral of two
> > densities over rectangular regions, using an adaptive integration
> > method. Other R functions compute densities and generate random
> > numbers.
> >
> > The starting motivation to write it was the  need to have functions
> > which compute the distribution functions in a non-Monte Carlo form,
> > sinse this caused me problems when these probabilities are involved in
> > a minimization problem. For this reason, I could not make use of the
> > CRAN package 'mvtnorm'.  Exactly to avoid superposition with the CRAN
> > package, 'mnormt' is made available somehere else, in case other
> > people want to use it.  The package 'mnormt' is at
> >    http://azzalini.stat.unipd.it/SN/Pkg-mnormt
> >
> > As explained, this is not uploaded to CRAN just to avoid clash with
> > the existing package. However, if it is felt appropriate, I have no
> > objection to upload it to CRAN.
> >
> > Best wishes,
> >
> > Adelchi Azzalini
>

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Reply | Threaded
Open this post in threaded view
|

Re: mutlivariate normal and t distributions

Adelchi Azzalini
On Tue, 24 Jan 2006 08:31:16 +0100 (CET), [hidden email] wrote:

TD>
TD>  yes, Adelchi, you should definitely submit the package to CRAN!

thanks for the feedback

I have just now uploaded the 'mnormt' package to CRAN, i.e.
 ftp.ci.tuwien.ac.at/incoming/
perhaps at a later stage we can find a way to
merge 'mnormt' and 'mvtnorm'

best wishes

Adelchi

--
Adelchi Azzalini  <[hidden email]>
Dipart.Scienze Statistiche, Università di Padova, Italia
tel. +39 049 8274147,  http://azzalini.stat.unipd.it/

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html