wfolsom/cs8803p4
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fixed unit tests, changed MDP generation to more reasonably seek the goal state, avoiding premature end of game.

Browse Source 
Removed unused google-code classes.
Regenerate policy when AdaptiveComPlayer.setTarget() is called.
This commit is contained in:
Woody Folsom
2012-04-30 17:37:37 -04:00
parent 3800436cd9
commit 8f92ae65d8
19 changed files with 53 additions and 939 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
test/aima/core/probability/mdp/PolicyIterationTest.java
View File

@@ -1,15 +1,8 @@
package aima.core.probability.mdp;
import java.util.Map;
import org.junit.Assert;
import org.junit.Before;
import org.junit.Test;
import aima.core.environment.cellworld.Cell;
import aima.core.environment.cellworld.CellWorld;
import aima.core.environment.cellworld.CellWorldAction;
import aima.core.environment.cellworld.CellWorldFactory;
import aima.core.environment.gridworld.GridCell;
import aima.core.environment.gridworld.GridWorld;
import aima.core.environment.gridworld.GridWorldAction;
@@ -18,7 +11,6 @@ import aima.core.probability.example.MDPFactory;
import aima.core.probability.mdp.MarkovDecisionProcess;
import aima.core.probability.mdp.impl.ModifiedPolicyEvaluation;
import aima.core.probability.mdp.search.PolicyIteration;
import aima.core.probability.mdp.search.ValueIteration;
/**
* @author Ravi Mohan
@@ -29,28 +21,31 @@ public class PolicyIterationTest {
public static final double DELTA_THRESHOLD = 1e-3;
private GridWorld<Double> gw = null;
private MarkovDecisionProcess<GridCell<Double>, GridWorldAction> mdp = null;
private MarkovDecisionProcess<GridCell<Double>, GridWorldAction> mdp = null;
private PolicyIteration<GridCell<Double>, GridWorldAction> pi = null;
final int maxTiles = 6;
final int maxScore = 10;
@Before
public void setUp() {
//take 10 turns to place 6 tiles
// take 10 turns to place 6 tiles
double defaultPenalty = -0.04;
gw = GridWorldFactory.createGridWorldForTileGame(maxTiles,maxScore,defaultPenalty);
gw = GridWorldFactory.createGridWorldForTileGame(maxTiles, maxScore,
defaultPenalty);
mdp = MDPFactory.createMDPForTileGame(gw, maxTiles, maxScore);
//gamma = 1.0
PolicyEvaluation<GridCell<Double>,GridWorldAction> pe = new ModifiedPolicyEvaluation<GridCell<Double>, GridWorldAction>(100,0.9);
// gamma = 1.0
PolicyEvaluation<GridCell<Double>, GridWorldAction> pe = new ModifiedPolicyEvaluation<GridCell<Double>, GridWorldAction>(
100, 0.9);
pi = new PolicyIteration<GridCell<Double>, GridWorldAction>(pe);
}
@Test
public void testPolicyIterationForTileGame() {
Policy<GridCell<Double>, GridWorldAction> policy = pi.policyIteration(mdp);
Policy<GridCell<Double>, GridWorldAction> policy = pi
.policyIteration(mdp);
for (int j = maxScore; j >= 1; j--) {
StringBuilder sb = new StringBuilder();
@@ -60,21 +55,5 @@ public class PolicyIterationTest {
}
System.out.println(sb.toString());
}
//Assert.assertEquals(0.705, U.get(gw.getCellAt(1, 1)), DELTA_THRESHOLD);
/*
Assert.assertEquals(0.762, U.get(cw1.getCellAt(1, 2)), DELTA_THRESHOLD);
Assert.assertEquals(0.812, U.get(cw1.getCellAt(1, 3)), DELTA_THRESHOLD);
Assert.assertEquals(0.655, U.get(cw1.getCellAt(2, 1)), DELTA_THRESHOLD);
Assert.assertEquals(0.868, U.get(cw1.getCellAt(2, 3)), DELTA_THRESHOLD);
Assert.assertEquals(0.611, U.get(cw1.getCellAt(3, 1)), DELTA_THRESHOLD);
Assert.assertEquals(0.660, U.get(cw1.getCellAt(3, 2)), DELTA_THRESHOLD);
Assert.assertEquals(0.918, U.get(cw1.getCellAt(3, 3)), DELTA_THRESHOLD);
Assert.assertEquals(0.388, U.get(cw1.getCellAt(4, 1)), DELTA_THRESHOLD);
Assert.assertEquals(-1.0, U.get(cw1.getCellAt(4, 2)), DELTA_THRESHOLD);
Assert.assertEquals(1.0, U.get(cw1.getCellAt(4, 3)), DELTA_THRESHOLD);*/
}
}