DataEvaluator.java

package com.datascience.gal.evaluation;

import com.datascience.core.base.LObject;
import com.datascience.core.nominal.NominalProject;
import com.datascience.core.nominal.decision.DecisionEngine;
import com.datascience.core.nominal.decision.IObjectLabelDecisionAlgorithm;
import com.datascience.core.nominal.decision.ObjectLabelDecisionAlgorithms;
import com.datascience.utils.ProbabilityDistributions;
import com.google.common.base.Strings;
import java.util.Collection;
import java.util.HashMap;
import java.util.Map;

import static com.datascience.core.nominal.ProbabilityDistributions.generateOneLabelDistribution;

/**
 *
 * @author konrad
 */
public class DataEvaluator {
	
	protected String labelChoosingMethod;
	protected IObjectLabelDecisionAlgorithm olda;

	public DataEvaluator(String method) {
		if (Strings.isNullOrEmpty(method)) {
			method = "soft";
		}
		labelChoosingMethod = method.toLowerCase();
		if (!labelChoosingMethod.equals("soft"))
			olda = ObjectLabelDecisionAlgorithms.get(labelChoosingMethod);
	}

	protected double evaluate(NominalProject project, LObject<String> datum) {
		Map<String, Double> dest_probabilities;
		if (labelChoosingMethod.equals("soft")) {
			dest_probabilities = project.getObjectResults(datum).getCategoryProbabilites();
		} else {
			dest_probabilities = generateOneLabelDistribution(
					project, new DecisionEngine(null, olda).predictLabel(project, datum));
		}
		double cost = 0.0;
		for (Map.Entry<String, Double> e : dest_probabilities.entrySet()) {
			Double misclassification_cost = project.getData().getCostMatrix().getCost(datum.getEvaluationLabel(), e.getKey());
			cost += e.getValue() * misclassification_cost;
		}
		return cost;
	}
	
	public Map<LObject<String>, Double> evaluate(NominalProject project){
		Collection<LObject<String>> evalData = project.getData().getEvaluationObjects();
		Map<LObject<String>, Double> ret = new HashMap<LObject<String>, Double>();
		for (LObject<String> cl: evalData) {
			ret.put(cl, evaluate(project, cl));
		}
		return ret;
	}
}