GibbsAskTest.java

package aima.test.core.unit.probability.bayes.approx;

import static org.junit.Assert.assertEquals;

import org.junit.Assert;
import org.junit.Test;

import aima.core.probability.CategoricalDistribution;
import aima.core.probability.RandomVariable;
import aima.core.probability.bayes.BayesianNetwork;
import aima.core.probability.bayes.approx.GibbsAsk;
import aima.core.probability.bayes.approx.LikelihoodWeighting;
import aima.core.probability.bayes.approx.RejectionSampling;
import aima.core.probability.bayes.impl.BayesNet;
import aima.core.probability.bayes.impl.FullCPTNode;
import aima.core.probability.domain.BooleanDomain;
import aima.core.probability.example.BayesNetExampleFactory;
import aima.core.probability.example.ExampleRV;
import aima.core.probability.proposition.AssignmentProposition;
import aima.core.probability.util.RandVar;
import aima.core.util.MockRandomizer;

/**
 * 
 * @author Ciaran O'Reilly
 * @author Ravi Mohan
 */
public class GibbsAskTest {
	public static final double DELTA_THRESHOLD = 0.1;

	/** Mock randomizer - A very skewed distribution results from the choice of 
	 * MockRandomizer that always favours one type of sample over others
	 */
	@Test
	public void testGibbsAsk_mock() {
		BayesianNetwork bn = BayesNetExampleFactory
				.constructCloudySprinklerRainWetGrassNetwork();
		AssignmentProposition[] e = new AssignmentProposition[] { new AssignmentProposition(
				ExampleRV.SPRINKLER_RV, Boolean.TRUE) };
		MockRandomizer r = new MockRandomizer(new double[] { 0.5, 0.5, 0.5,
				0.5, 0.5, 0.5, 0.6, 0.5, 0.5, 0.6, 0.5, 0.5 });

		GibbsAsk ga = new GibbsAsk(r);

		double[] estimate = ga.gibbsAsk(
				new RandomVariable[] { ExampleRV.RAIN_RV }, e, bn, 1000)
				.getValues();

		Assert.assertArrayEquals(new double[] { 0, 1 }, estimate, DELTA_THRESHOLD);
	}
	
	/** Same test as above but with JavaRandomizer
	 * <p>
	 * Expected result : <br/>
	 * P(Rain = true | Sprinkler = true) = 0.3 <br/>
	 * P(Rain = false | Sprinkler = true) = 0.7 <br/>
	 */
	@Test
	public void testGibbsAsk_basic() {
		BayesianNetwork bn = BayesNetExampleFactory
				.constructCloudySprinklerRainWetGrassNetwork();
		AssignmentProposition[] e = new AssignmentProposition[] { new AssignmentProposition(
				ExampleRV.SPRINKLER_RV, Boolean.TRUE) };
		
		GibbsAsk ga = new GibbsAsk();

		double[] estimate = ga.gibbsAsk(
				new RandomVariable[] { ExampleRV.RAIN_RV }, e, bn, 1000)
				.getValues();

		Assert.assertArrayEquals(new double[] { 0.3, 0.7 }, estimate, DELTA_THRESHOLD);
	}
	
	@Test
	public void testGibbsAsk_compare() {
		// create two nodes: parent and child with an arc from parent to child 
	    RandomVariable rvParent = new RandVar("Parent", new BooleanDomain());
	    RandomVariable rvChild = new RandVar("Child", new BooleanDomain());
	    FullCPTNode nodeParent = new FullCPTNode(rvParent, new double[] { 0.7, 0.3 });
	    new FullCPTNode(rvChild, new double[] { 0.8, 0.2, 0.2, 0.8 }, nodeParent);

	    // create net
	    BayesNet net = new BayesNet(nodeParent);

	    // query parent probability
	    RandomVariable[] rvX = new RandomVariable[] { rvParent };

	    // ...given child evidence (true)
	    AssignmentProposition[] propE = new AssignmentProposition[] { new AssignmentProposition(rvChild, Boolean.TRUE) };

	    // sample with LikelihoodWeighting
	    CategoricalDistribution samplesLW = new LikelihoodWeighting().ask(rvX, propE, net, 1000);
	    assertEquals(0.9, samplesLW.getValue(Boolean.TRUE), DELTA_THRESHOLD);

	    // sample with RejectionSampling
	    CategoricalDistribution samplesRS = new RejectionSampling().ask(rvX, propE, net, 1000);
	    assertEquals(0.9, samplesRS.getValue(Boolean.TRUE), DELTA_THRESHOLD);

	    // sample with GibbsAsk
	    CategoricalDistribution samplesGibbs = new GibbsAsk().ask(rvX, propE, net, 1000);
	    assertEquals(0.9, samplesGibbs.getValue(Boolean.TRUE), DELTA_THRESHOLD);
	}
}