TracNav?

PhaseLibTuner

The PhaseLibTuner is to provide a basic and easy to use interface to existing algorithms. The prototype allows the quick selection of an algorithm and to immediately run and test it with a given set of parameters.

Overview

The Tuner is started by running (the main method of) the class de.dfki.km.phaselib.tools.tuner.Tuner. The UI basically consits of an upper and a lower half. In the upper one, the algorithm to be used is to be specified. Therefore, the algorithm class is selected, and the set of SimilarityMeasures to be employed with it. On the right hand side, parameters can be set, to influence the algorithms behaviour. The most common parameters are already entered in the table, but you can enter further parameters at will. Please note, that the parameters, as well as the set of similarity measures will only come into account, if the selected algorithm supports them. So does, for example, the SingleEvidenceAlignment algorithm only use the first selected SimilarityMeasure for its calculations.

In the lower part of the UI the developer may specify, what to do with the selected algorithm. The architecture is designed to host several plugable tabs, but by now only one such tab exists.

Tab: Single Benchmark Evaluation

This tab uses the selected algorithm to execute it on a single benchmark of the INRIA set. The algorithm is run and immediately evaluated. The results are kept in a table beneath the benchmark selector. On request, the alignments produced so far can be viewed and compared using the VizAlign visualisation system. There is also a console that should catch most of the output.

Requirements

The PhaseLibTuner tries to find all available instances of AlignmentGenerator and SimilarityMeasure upon startup. To be able to use these, the following conditions should be met:

• An AlignmentGenerator should be in de.dfki.km.phaselib.impl.alignments or its subpackages and the classfile must be found in the class path. Similarly, any SimilarityMeasure has to be in de.dfki.km.phaselib.impl.similarities or beneath.
• An AlignmentGenerator must provide:
  • a default (empty) constructor or
  • a constructor taking a single SimilarityMeasure or
  • a constructor taking a Collection or Array of SimilarityMeasures
• A SimilarityMeasure must provide a default (empty) constructor.

If the constructor conditions are not met, the module may be selectible but will fail on initialisation.