Evaluating alignments

On top of the alignment API, there is an evaluation API and an implementation described here. It is included in the Alignment toolkit

Evaluating

For demonstrating the use of the API, we implement it and use it through a particular processor (fr.inrialpes.exmo.align.util.EvalAlign) which:

Running the program is achieved through:
$ java -cp lib/procalign.jar fr.inrialpes.exmo.align.util.EvalAlign
Require two alignement filenames
usage: EvalAlign [options] file1 file2
options are:
        --debug[=n] -d[=n]              Report debug info at level ,
        --impl[=classname] -i classname Use classname for implementing Evaluator
        --help -h                       Print this message
or ($CWD is the current directory)
$ java -cp lib/procalign.jar fr.inrialpes.exmo.align.util.EvalAlign file://$CWD/result/align1.owl file://$CWD/result/align2.owl 

Implementing and extending the API

A (default) impplementation of this API can be found in the fr.inrialpes.exmo.align.impl package.

Adding new evaluations methods amounts to create a new Evaluator class implementing the interface. Generally, this can extend the proposed fr.inrialpes.exmo.align.impl.BasicEvaluator class.

The BasicEvaluator has the good taste of defining the storage sructures for alignment, but that is all.

The BasicEvaluator has been extended so as to provide different evaluation:

PRecEvaluator
computes Precision/Recall rates as well as derived measures.
ExtPRecEvaluator
provide the relaxed precision and recall as defined in a paper by Ehrig and Euzenat in 2005.


http://alignapi.gforge.inria.fr/align.html

$Id: eval.html 469 2007-04-25 08:44:06Z euzenat $