Re: [DL-Learner discussion] How to initialize FIC one time?

[Jens L. <le...@in...>]

Hello,

On 27.03.2012 20:36, Lorenz Bühmann wrote:
> Hi,
> 
> using several learning problems in one file is not possible, but it is
> possible to define several learning algorithms with the latest SVN
> version. 

To make it a bit more clearer: The command line interface will run all
algorithms declared in the conf file (in the latest SVN version).
However, the algorithms currently do not support several learning
algorithms as input. That means that something like this would work:

// knowledge source definition
ks.type = "OWL File"
ks.fileName = "example.owl"

// reasoner
reasoner.type = "fast instance checker"
reasoner.sources = { ks }

// learning problem
lp.type = "posNegStandard"
lp.positiveExamples = { ... }
lp.negativeExamples = { ... }

// learning problem 2
lp2.type = "posNegStandard"
lp2.positiveExamples = { ... }
lp2.negativeExamples = { ... }

// create learning algorithms to run problems
alg.type = "celoe"
alg.learningProblem = lp
alg.maxExecutionTimeInSeconds = 1

alg2.type = "celoe"
alg2.learningProblem = lp2
alg2.maxExecutionTimeInSeconds = 1

Here, you define two learning problems and separate algorithms for
running them. Both re-use the reasoner object. Of course, a drawback of
this method is that the conf files could become very verbose in case you
configure a lot of algorithm parameters. In such cases, a programmatic
solution may be simpler as Lorenz pointed out.

(Note that you need the SVN version for this, which also fixes the
problems with ELTL reported in your other post.)

Kind regards,

Jens

-- 
Dr. Jens Lehmann
AKSW/MOLE Group, Department of Computer Science, University of Leipzig
Homepage: http://www.jens-lehmann.org
GPG Key: http://jens-lehmann.org/jens_lehmann.asc