Examples of org.jamesii.asf.spdm.generators.IPerformancePredictor

org.jamesii.asf.spdm.generators.IPerformancePredictor
Interface for all performance predictors. These are components that predict the performance of an (algorithmic) configuration for a given problem, which is characterized by certain features. More formally, the interface represents functions of the form F x A -> R, which assign a certain weight for each algorithm in A to be applied to a problem with features in F. The weight may be interpreted in different ways, generally is should refer to the eligibility and suitability of an algorithm, i.e. a return value of 0 indicates that the algorithm is not applicable to the problem (the higher the weight, the better). This allows to use performance predictors for a broad variety of algorithm synthesis/selection tasks, i.e. for picking a single algorithm from a set or for defining an algorithm portfolio. @author Roland Ewald

    IDataSelector trainingDataSelector =
        new SimpleDataSelector(0, trainEndIndex);
    List<PerformanceTuple> trainingData = trainingDataSelector.selectData(data);
    IPerformancePredictorGenerator predGen =
        predGenFactory.createPredictorGenerator(parameters, data.get(0));
    IPerformancePredictor predictor =
        predGen.generatePredictor(trainingData, dataSet.getMetaData());


    // Evaluate predictor
    IDataSelector testDataSelector =
        new SimpleDataSelector(trainEndIndex, data.size());

View Full Code Here

      SimSystem.report(
          Level.INFO,
          "\tCreating predictor with '"
              + Strings.dispClassName(factory.getClass()) + " / "
              + parameters.toString() + "'...");
      IPerformancePredictor predictor =
          createPredictor(factory, dataSet, parameters, trainingData);
      SimSystem.report(Level.INFO, "done.\n\tStarting evaluation...");
      results.add(predictorEvaluator.evaluate(predictor, trainingData,
          testData, parameters));
      SimSystem.report(Level.INFO, "done.");

View Full Code Here

  @Override
  public double predictPerformance(Features features, Configuration config) {
    if (!predictors.containsKey(config)) {
      return Double.POSITIVE_INFINITY;
    }
    IPerformancePredictor predictor = predictors.get(config);
    return predictor.predictPerformance(features, config);
  }

View Full Code Here


      // Create predictor generator, predictor
      IPerformancePredictorGenerator predictorGenerator =
          predGenFactory.createPredictorGenerator(predGenParameters,
              inputData.get(0));
      IPerformancePredictor predictor =
          predictorGenerator.generatePredictor(inputData, featureMData);
      predictors.put(configTuples.getKey(), predictor);
    }


    return new EnsemblePerfPredictor(predictors);

View Full Code Here

   *           when predictor generation goes wrong
   */
  public void testPredictorGeneration() throws Exception {


    // Generate predictor
    IPerformancePredictor predictor =
        predictorGenerator.generatePredictor(getDataSet().getInstances(),
            getDataSet().getMetaData());
    assertNotNull(predictor);


    // Generate corresponding selector
    ISelector selector =
        SelectorGeneration.createSelector(getDataSet(), predictor);
    assertNotNull(selector);


    // Simple test run to see if comparison throws Exceptions etc.
    Features features = getDataSet().getInstances().get(0).getFeatures();
    Configuration config1 =
        getDataSet().getInstances().get(0).getConfiguration();
    Configuration config2 =
        getDataSet().getInstances().get(1).getConfiguration();


    try {
      predictor.predictPerformance(features, config1);
      predictor.predictPerformance(features, config2);
    } catch (Exception e) {
      SimSystem.report(e);
      fail("Performance prediction should work.");
    }
  }

View Full Code Here

    List<PerformanceTuple> data = dataSet.getInstances();
    Pair<List<PerformanceTuple>, List<PerformanceTuple>> bootstrapData =
        bootStrap(data);
    IPerformancePredictorGenerator predGen =
        predGenFactory.createPredictorGenerator(parameters, data.get(0));
    IPerformancePredictor predictor =
        predGen.generatePredictor(bootstrapData.getFirstValue(),
            dataSet.getMetaData());
    IPredictorEvaluator sev = new FullPredictorEvaluator();
    PredictorPerformance selPerf =
        sev.evaluate(predictor, bootstrapData.getFirstValue(),

View Full Code Here

TOP

Related Classes of org.jamesii.asf.spdm.generators.IPerformancePredictor

org.jamesii.asf.spdm.generators.ensemble.EnsemblePerfPredictor

org.jamesii.asf.spdm.generators.ensemble.EnsemblePredictorGenerator

org.jamesii.simspex.spdm.evaluation.bootstrapping.BootStrapping

org.jamesii.simspex.spdm.evaluation.crossvalidation.CrossValidation

org.jamesii.simspex.spdm.evaluation.simple.SimpleEvaluationStrategy

org.jamesii.simspex.spdm.generators.PredictorGeneratorTest

All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.