Re: Debian-AI

[Andrew Dougherty <andrewdo@frdcsa.org>]

From: Isabel Drost <mainec@isabel-drost.de>
Subject: Re: Debian-AI
Date: Wed, 2 Jan 2008 11:17:28 +0100

> If your goal is to provide a comprehensive repository of AI software, I think 
> you have forgotten some rather important AI projects on your list. You might 
> want to have a look at the following two lists:
> 
> http://ml-site.grantingersoll.com/index.php?title=Existing_Learning_Tools
> 
> http://mloss.org/software/

The software Isabel Drost has mentioned is an example of the thousands
of high-end systems for which no packages exist.

The website http://frdcsa.onshore.net/frdcsa/external is really a list
of software that has been acquired and packages attempted.  The list
of all known software (software ontology) has not been made, see [0]
for a description of the attempt so far.

The issue of "automatic packaging" is IMHO a red-herring.  Even if the
rough packages are done completely by hand, the problem remains that
there is a lot of useful, properly licensed software for which no
package exists.  Consequently, the software has not been mapped to the
FHS properly, it is more difficult to develop additional software
which depends upon it, it is more difficult for any user to install
the software, etc.  The benefits of packaging are great.  It promotes
code reuse, and also makes it easier to discover functionality.

Any ideas as to how we can make all these packages?  Is there any
interest in some kind of effort aimed at this, to increase the package
coverage?  It's not possible to spam WNPP, nor create all the packages
myself.  Can Debian create some kind of official USE AT YOUR OWN RISK
repository of user-contributed packages?  Would people like to work on
making rough packages from the list?  Would they like to write tools
that reduce rough package creation to a wizard, so that Linux users
with no Debian expertise with an interest in having packages can
create them (and link to this tool from Sourceforge and Freshmeat
saying: "Want a package of this?").  Can we start a business where we
contract to make packages for people?  Can we build mappings between
packages in Debian and other distributions and automatically convert
these using Alien plus some tools to correct dependencies?  Would
anyone like to join my project to create these packages, or give me
advice how to go about creating them?

I've contacted http://debian.pkgs.cpan.org/ and expect to help with
that effort.  Thank you very much for the link.

Kind regards,
Andrew Dougherty

P.S. The packages of all FRDCSA internal codebases are being tweaked
for release, so these may be available within the next few weeks, if
anyone is interested.

[0] The CSO system (http://frdcsa.onshore.net/frdcsa/internal/cso) has
a MySQL database of all the software from Freshmeat, Sourceforge, and
several other major software indexes, data from the FLOSSMole project.
(linking redundant entries can be done using the (non-free) MNOP tool
(http://www.autonlab.org/autonweb/10514), but they've not yielded
their software.)  The problem is similar to the problem of "Web People
Search"
http://nlp.cs.swarthmore.edu/semeval/tasks/task13/summary.shtml.

The process of finding and inserting metasites such as the ones Isabel
Drost has mentioned is more complex.  An information extraction
technique called MDR (http://citeseer.ist.psu.edu/liu03mining.html)
has been mostly implemented for extracting software from metasites,
and automatic detection of metasites using one of the projects called
WebKB (http://www.cs.cmu.edu/~webkb/) has been experimented with.  The
most practical method is to just do it semi-automatically using the
RADAR system (http://frdcsa.onshore.net/frdcsa/internal/radar).  But
the best method would be for people to submit metasites to a
repository and then either automatically or semiautomatically process
those.

Once the packages are found, their capabilities need to be
indexed/formalized.  This task is beyond the scope of this letter.
Most of the tools which are necessary to accomplish this are however
not packaged.