Examples of cc.mallet.types.FeatureInducer

cc.mallet.types.FeatureInducer

  // But it is here as a reminder to do something about induceFeaturesFor(). */
  @Deprecated
  public Sequence[] predict (InstanceList testing) {
    testing.setFeatureSelection(this.globalFeatureSelection);
    for (int i = 0; i < featureInducers.size(); i++) {
      FeatureInducer klfi = (FeatureInducer)featureInducers.get(i);
      klfi.induceFeaturesFor (testing, false, false);
    }
    Sequence[] ret = new Sequence[testing.size()];
    for (int i = 0; i < testing.size(); i++) {
      Instance instance = testing.get(i);
      Sequence input = (Sequence) instance.getData();

View Full Code Here

  /** When the CRF has done feature induction, these new feature conjunctions must be 
   * created in the test or validation data in order for them to take effect. */
  public void induceFeaturesFor (InstanceList instances) {
    instances.setFeatureSelection(this.globalFeatureSelection);
    for (int i = 0; i < featureInducers.size(); i++) {
      FeatureInducer klfi = featureInducers.get(i);
      klfi.induceFeaturesFor (instances, false, false);
    }
  }

View Full Code Here

              gainFactory = new ExpGain.Factory (lvs, gaussianPriorVariance);
            else if (gainName.equals("grad"))
              gainFactory =  new GradientGain.Factory (lvs);
            else if (gainName.equals("info"))
              gainFactory =  new InfoGain.Factory ();
            klfi[i][j] = new FeatureInducer (gainFactory,
                clusteredErrorInstances[i][j], 
                numFeaturesPerFeatureInduction,
                2*numFeaturesPerFeatureInduction,
                2*numFeaturesPerFeatureInduction);
            crf.featureInducers.add(klfi[i][j]);
          }
        }
        for (int i = 0; i < numLabels; i++) {
          for (int j = 0; j < numLabels; j++) {
            logger.info ("Adding new induced features for "+
                crf.outputAlphabet.lookupObject(i)+" -> "+crf.outputAlphabet.lookupObject(j));
            if (klfi[i][j] == null) {
              logger.info ("...skipping because no features induced.");
              continue;
            }
            // Note that this adds features globally, but not on a per-transition basis
            klfi[i][j].induceFeaturesFor (trainingData, false, false);
            if (testingData != null) klfi[i][j].induceFeaturesFor (testingData, false, false);
          }
        }
        klfi = null;
      } else {
        int s = errorLabelVectors.size();
        LabelVector[] lvs = new LabelVector[s];
        for (int i = 0; i < s; i++)
          lvs[i] = (LabelVector) errorLabelVectors.get(i);


        RankedFeatureVector.Factory gainFactory = null;
        if (gainName.equals ("exp"))
          gainFactory = new ExpGain.Factory (lvs, gaussianPriorVariance);
        else if (gainName.equals("grad"))
          gainFactory =  new GradientGain.Factory (lvs);
        else if (gainName.equals("info"))
          gainFactory =  new InfoGain.Factory ();
        FeatureInducer klfi =
          new FeatureInducer (gainFactory,
              errorInstances, 
              numFeaturesPerFeatureInduction,
              2*numFeaturesPerFeatureInduction,
              2*numFeaturesPerFeatureInduction);
        crf.featureInducers.add(klfi);
        // Note that this adds features globally, but not on a per-transition basis
        klfi.induceFeaturesFor (trainingData, false, false);
        if (testingData != null) klfi.induceFeaturesFor (testingData, false, false);
        logger.info ("CRF4 FeatureSelection now includes "+crf.globalFeatureSelection.cardinality()+" features");
        klfi = null;
      }
      // This is done in CRF4.train() anyway
      //this.setWeightsDimensionAsIn (trainingData);

View Full Code Here

  // But it is here as a reminder to do something about induceFeaturesFor(). */
  @Deprecated
  public Sequence[] predict (InstanceList testing) {
    testing.setFeatureSelection(this.globalFeatureSelection);
    for (int i = 0; i < featureInducers.size(); i++) {
      FeatureInducer klfi = (FeatureInducer)featureInducers.get(i);
      klfi.induceFeaturesFor (testing, false, false);
    }
    Sequence[] ret = new Sequence[testing.size()];
    for (int i = 0; i < testing.size(); i++) {
      Instance instance = testing.get(i);
      Sequence input = (Sequence) instance.getData();

View Full Code Here

  /** When the CRF has done feature induction, these new feature conjunctions must be 
   * created in the test or validation data in order for them to take effect. */
  public void induceFeaturesFor (InstanceList instances) {
    instances.setFeatureSelection(this.globalFeatureSelection);
    for (int i = 0; i < featureInducers.size(); i++) {
      FeatureInducer klfi = featureInducers.get(i);
      klfi.induceFeaturesFor (instances, false, false);
    }
  }

View Full Code Here

TOP

Related Classes of cc.mallet.types.FeatureInducer

cc.mallet.fst.CRF

cc.mallet.fst.CRFTrainerByLabelLikelihood

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.