[Debtags-devel] AI Tagger

Benjamin Mesing bensmail at gmx.net
Tue Aug 16 19:09:43 UTC 2005


Hello

> That is nice.  Or the other way round: if I adopt the use::editing
> facet, it could pop me up names of packages that could use or not use
> that tag.
Ok, this sounds useful and I will try to implement this as soon as I
have some time at hands (which might take some time -- packagesearch
waits for conversion to QT4 too...).

> About giving it a try, I'd like to.  What I'd like to do is to run it on
> the commandline, get a tag patch out, review it in the new debtags-edit
> patch reviewer[1] and commit what's good of it.
Yeah, this sounds like a good integration. Do you have a specification
for the patch system at hand? Or should I use the  perl interface of
libdebtags (which will probably mean more effort, but perhaps I should
switch to use apt-front anyways...)? 


> I understand that so far I can only do it one tag at a time, and that'd
> be allright.  I understand I should choose a tag (say, use::editing),
> then train the engine, then run it to get the likely packages.
Totally correct, however "then run it to get the likely packages" is not
yet implemented -- but will be implemented soon (see above).

> Can you show me the two example commandlines?  I can take care of
> generating the tag patch from the output.
Ok, here is what you can do right now:
     1. ./create-data.pl use::editing --max-good=100 --bad-ratio=2
     2. ./bayesian-tagger.pl use::editing
     3. ./bayesian-tagger.pl --test-package kwrite
The first step will create training and testdata. 
The second will train and test with the created data, and print some
statistics. 
The third will test the package kwrite.

You can check out the bayesian tagger at:
        svn.debian.org/svn/debtags/autodebtag/trunk/ai-tagger

OK, have to go to bed know, will answer the rest tomorrow.

Greetings Ben






More information about the Debtags-devel mailing list