neue Rmetrics packages

[Diethelm Wuertz]

Dirk Eddelbuettel wrote:
> Diethelm,
>
> Thanks for the reply.  Let me prefix my follow-up by stressing that I am
> wearing two hats here, and that can be confusing
>
> a) as someone interested in seeing Rmetrics prosper:  I think the new
> packages are the right idea, as I had hinted earlier when we were about to
> set up the list.
>
> b) as the acting Debian maintainer for the bundle, I am unhappy about the
> changes and the lack of a simple heads-up.  When I add new packages to
> Debian, it takes two to three weeks for them to be approved and available.
> So only telling me on release date that 'foo, bar and bling are now needed'
> is suboptimal as I could have used that information three weeks prior ...
>
> That said, the rest of the mail will be with the Debian hat:
>
> | Is somebody around who can interface the interpolation functions from 
> | the GNU scientific library to R? 
>
> Okay, I volunteer to do this. I have been the Debian maintainer for the GSL
> and know bits of the code base, and I also successfully ported code out of
> the GSL into RDieharder.  So let's work on this together and take the threat
> off-line:  can you help me with pointers about what calls Rmetrics makes to
> akima, what GSL functions you had in mind etc pp?
>
> That way we get rid of akima.
>   
I'm no longer sure if GSL has a irregular bivariate interpolation 
function (which also shape preserving)  ...
I have to look in more detail into this which needs some time ...
> | There is another problem with the package fPortfolio. It uses currently the
> | nonlinear solver with nonlinear constraints Rdonlp2 - it comes as a builtin
> | with the package.
> | 
> | http://arumat.net/Rdonlp2/
> | 
> | I think this has also to be replaced by another GNU or GNU like licensed 
> | solver. A possible candidate may be
> | 
> | http://www.coin-or.org/Bonmin/Intro/index.html
>
> That is harder.
>
> | Another question - is there a reason why robustbase is not part of Debian ?
>
> Lack of manpower?  But I agree -- I will add RUnit and robustbase to
> Debian. Both are 'good to have'.
>
>   

> That leaves two more:  sn, and the mnormt package that sn depends upon.
> Could we possibly alter Rmetrics to make this a run-time test and a Suggests
> rather than a Depends?
>   
There was several times the discussion to remove Builtins from Rmetrics -
so I removed sn and mvtnorm, but at that time I didn't realize that they are
not available under Debian.

What would be the best solution ? To add it as Suggests - or again to 
copy that
part of the code as Builtin back  to Rmetrics ?  Missing of sn and 
mvtnorm will
have a  severe influence on Rmetrics, since anything what has to do with
multivarate assets and portfolios depends more or less on sn and mvtnorm.
Any suggestions how we should proceed ?

(Nota bene, the same is with robustbase, since many robust methods may 
be used
in portfolio design. -- but you suggest it to add it to Debian.)
> Dirk
>
>   
Again, thank you Dirk for your quick reply,

I remember that we had already about two years ago the problem with
akima. So I was yesterday night looking again into the code of
fUtilities concerning the point with akima.

First there is NO akima Fortran/C code in fUtilities. Currently, the R
functions were written by my own along the R Code of akima which is GNU
licensed. You find the following lines and similar lines in the code of
these functions:

.Fortran(..., PACKAGE = "akima")

So doing "irregular spaced bivariate smooth interpolation" with Rmetrics
you will get an error that akima is missing. So the people have to add
and load the akima package by their own.

Has anybody a better solution to this problem?
So fUtilities can go to my opinion without any problems to Debian ...

Concerning fPortfolio, I will do the same procedure. There I will remove
donlp2 and work in the same way with  .Fortran(..., PACKAGE = "Rdonlp2").
Thus when using the donlp2 solver in some Portfolio optimization functions
one has to add the Rdonlp2 package, note no licensed code other than GNU
or GNU like will be part of Rmetrics.

Sorry, that I have created the confusion with akima, but the Rmetrics
code is meanwhile so huge, that I simply forgot the "akima" situation
for a moment.


DW