Re: Role of SHOGUN in Debian Med
Hi Soeren,
On Thu, Jan 24, 2013 at 10:39:54PM +0100, Soeren Sonnenburg wrote:
> the issue with shogun & debian currently is that it requires ~3.5GB of
> memory to compile for some of its interfaces. Build bots where choking
> on this so I have no idea what to do about it. That is the reason I gave
> up getting a current version into wheezy. So it is not a good idea to
> more forcefully depend on this package currently.
There is no thing in Blends metapackages like "forcefully Depends". In
metapackages only Recommends are used and only in the case that a
package is available in testing. That's technically verified at
metapackage creation time. So if there are good reasons to have some
package not in the current testing (as you mentioned) it will never be
recommended by the metapackage but rather suggested. Regarding the
suggests on the other hand we go far beyond the official package pool of
Debian and we do suggest packages in non-free or even not (yet) existent
in Debian. On one hand this makes sense to not always need to update
the metapackages if a package we are working on enters the Debian mirror
and the metapackage can also be used in derivatives with additional
packages.
So the relevant question whether a package should be mentioned in a
certain task is always:
Is a package relevant in the workfield of the task?
Yes -> Recommends
May be -> Suggests
No -> Not mentioned in the tasks file
> IMHO the most useful is the python_modular interface but surely it
> depends on which language one is used to / wants to use. I don't
> recommend to use any of the static interfaces nowadays (very limited
> feature subset / functionality).
Considering the explanation above would you agree that it makes sense to
put a
Suggests: shogun-python-modular
into bio-dev which means even if the package would be available in
testing the "Suggests" remains as a slight hint to the user that this
package might be interesting even if it is not installed without forcing
installation of suggests?
IMHO similar questions should be asked for the Debian Science tasks[1]
if you seem some task where some binray shogun package might fit here.
Finally you are doing some good work in developing the software and
distributing it via Debian. You should advertise this work to your
potential users and the Blends approach is one way to do so. I just
stumbled upon the package by pure chance even if I'm checking the Debian
package pool very thorough all the time.
> It certainly is one of the more popular ML toolkits. Back in academia I
> did a couple of tools with it that might still be state-of-the-art, like
> asp (a splice site detector) and arts (a human tss detector).
Sorry, just for the sake of interest: What is a tss detector? BTW,
this reminds me: It might not harm if you would try to enhance the long
description of the Shogun packages a bit to enable users who are not
that deeply involved into this field to draw some better conclusions.
Kind regards
Andreas.
> On Wed, 2013-01-23 at 14:32 +0100, Andreas Tille wrote:
> >
> > I'm also not sure whether we should apply a stronger relation
> > (=Recommends) or if SHOGUN is rather scratching a borderline field in
> > bioinformatics and thus should be rather only suggested. Please be
> > so kind and visit
> >
> > http://debian-med.alioth.debian.org/tasks/bio-dev and
> > http://debian-med.alioth.debian.org/tasks/bio
> >
> > to see what is currently listed there to give a reasonable
> > recommendation. Sorry if my question sounds a bit naive but we try to
> > assemble all packages relevant for bioinformatics and sometimes the
> > needed knowledge to draw a proper decision needs to be gathered
> > carefully.
[1] http://blends.alioth.debian.org/science/tasks/
--
http://fam-tille.de
Reply to: