Re: Regarding r-cran-rcppmlpack
Howdy,
On 31 January 2021 at 16:08, Nilesh Patra wrote:
| The new release of r-cran-genieclust seems to need r-cran-rcppmlpack (as seen here[1]) which is still unpackaged.
| Andreas did initial packaging of r-cran-rcppmlpack[2] (this has not yet been uploaded) - it builds and works good, however I quickly realised that it uses an "embedded" copy of mlpack library which is bad for obvious reasons.
| I tried repacking it, but then it seems that the embedded MLpack is several versions behind the mlpack we have in Debian.
As I wrote in the thread that you may have seen RcppMLPACK is outdated. It's
main upstream repo and CRAN package are
https://github.com/rcppmlpack/RcppMLPACK1
https://cloud.r-project.org/web/packages/RcppMLPACK/index.html
A careful glance reveals that all there is six years old. So I should just talk to
Qiang about it. We may just have to pull it from CRAN. Or keep maintaining it
as status quo.
As for your current inability to package it: I forget the CRAN / R details
but at some point R switched to this 'pivoted' build where it builds in
location A, pivots to location B and retests. That hits me in two or so CRAN
packages I co-author---and one generally gets a pass as "grandfathered" on
CRAN. However, I am not sure we _can_ in fact overcome that with Debian
packaging but I have not tried. So you should really consider retiring it.
And after *many* years and many failed attempts, the *current* MLPACK sources
are packaged at CRAN, and will remain packaged and update there as R is now
a formal primary delivery language in the source repo:
https://github.com/mlpack/mlpack
http://cloud.r-project.org/package=mlpack
You will note that they were kind enough to list me as an author/contributor
in their R package as I assisted over the years with the trials (included the
"v2" approach you link to below and which still failed). One very, very eager
and capable GSoC student pushed it far enough for all this to work.
That is what should get packaged for Debian. It's clean (but large so there
could be issues with some autobuilders -- but it's still way smaller than
some other things we have in Debian).
| And then, I saw that you have a fork of r-cran-rcppmlpack here[3] however the readme states that it is still experimental so I'm not sure if I should package this and upload for unstable.
| Or with this if genieclust can even work for that matter. I'm not sure if uploading the mlpack1 compatible package with embedded copy would be a good idea. (I don't feel so)
|
| Since you seem to be (mplack-3-compatible) r-cran-rcppmlpack's primary author, I'd really appreciate your advice on this - as to what do you think about its compatibility and feasibility to enter
| unstable and therefore testing.
No please see
https://github.com/mlpack/mlpack
http://cloud.r-project.org/package=mlpack
As such the whole rcppmlpack org is a little outdated. I will make a point of
upding the README.md files at GitHub today.
Thank you for pursuing packaging various forms of mlpack. It is worth it!
Dirk
| [1]: https://salsa.debian.org/r-pkg-team/r-cran-genieclust/-/blob/master/DESCRIPTION#L38
| [2]: https://salsa.debian.org/r-pkg-team/r-cran-rcppmlpack
| [3]: https://github.com/rcppmlpack/rcppmlpack2
|
| Nilesh
|
--
https://dirk.eddelbuettel.com | @eddelbuettel | edd@debian.org
Reply to: