StackedClassifierTeacher.java

package edu.cmu.minorthird.classify;

import edu.cmu.minorthird.classify.relational.*;
import java.util.*;

/**
 * Implements the teacher's side of the learner-teacher protocol for SGM.
 *
 * @author Zhenzhen Kou
 *
 */
public abstract class StackedClassifierTeacher extends ClassifierTeacher{

	/** Train a StackedGraphicalLearner and return the learned Classifier, using
	 * some unspecified source of information to get labels.  
	 */

	final public Classifier trainStacked(StackedBatchClassifierLearner learner){
		
		// initialize the learner for a new problem
		learner.reset();

		// tell learner the schema of examples
		learner.setSchema(schema());
		
		// I am commenting out the two lines below because they don't do anything - frank
		// learner.RelDataset.setAggregators(this.getAggregators());
		// learner.RelDataset.setLinksMap(this.getLinksMap());
		
		// provide unlabeled examples to the learner, for unsupervised
		// training, semi-supervised training, or active learner
		learner.setInstancePool(instancePool());

		// passive learning from already-available labeled data
		for(Iterator<Example> i=examplePool();i.hasNext();){
			learner.addExample(i.next());
		}
		// active learning 
		while(learner.hasNextQuery()&&hasAnswers()){
			Instance query=learner.nextQuery();
			Example answeredQuery=labelInstance(query);
			if(answeredQuery!=null){
				learner.addExample(answeredQuery);
			}
		}
		// signal that there's no more data available
		learner.completeTraining();

		// final result
		return learner.getClassifier();
	}

	/** The linkMaps for stacked graphical learning
	 */
	abstract protected Map<String,Map<String,Set<String>>> getLinksMap();

	/** The Aggregators for stacked graphical learning
	 */
	abstract protected Map<String,Set<String>> getAggregators();

}