Merge pull request #72 from dityas/dev

Dev
dityas · Mar 26, 2020 · 8737f4b · 8737f4b
2 parents 37a6a04 + 6daa3e0
commit 8737f4b
Show file tree

Hide file tree

Showing 45 changed files with 2,488 additions and 1,165 deletions.
diff --git a/Protos/build/Protos.jar b/Protos/build/Protos.jar
diff --git a/Protos/domains/tiger.L1.txt b/Protos/domains/tiger.L1.txt
@@ -268,7 +268,7 @@ discount  0.9
 tolerance 0.001
 
 (frames
-	(frame 0 level 0
+	(frame 0 level 0 def
 		(variables 
 		 	(tiger-location tiger-left tiger-right))
 

diff --git a/Protos/domains/tiger.L1multiple_new_parser.txt b/Protos/domains/tiger.L1multiple_new_parser.txt
@@ -148,7 +148,7 @@ discount  0.9
 tolerance 0.001
 
 (frames
-	(frame 0 level 0
+	(frame 0 level 0 def
 		(variables 
 		 	(tiger-location tiger-left tiger-right))
 
@@ -197,7 +197,7 @@ tolerance 0.001
 		tolerance 0.001
 	)
 
-	(frame 1 level 0
+	(frame 1 level 0 def
 		(variables 
 		 	(tiger-location tiger-left tiger-right))
 

diff --git a/Protos/src/thinclab/belief/SSGABeliefExpansion.java b/Protos/src/thinclab/belief/SSGABeliefExpansion.java
@@ -8,7 +8,6 @@
 package thinclab.belief;
 
 import java.util.ArrayList;
-import java.util.Arrays;
 import java.util.HashSet;
 import java.util.List;
 
@@ -49,6 +48,7 @@ public class SSGABeliefExpansion extends BeliefRegionExpansionStrategy {
 
 	/* number of iterations of SSGA during each expansion */
 	private int nIterations;
+	private int maxBeliefs = 500;
 
 	private static final Logger LOGGER = Logger.getLogger(SSGABeliefExpansion.class);
 
@@ -81,6 +81,7 @@ public SSGABeliefExpansion(POMDP p, int maxDepth, int iterations) {
 		this.exploredBeliefs.addAll(fb.getBeliefPoints());
 
 		fb = null;
+		this.maxBeliefs = 500;
 
 		LOGGER.debug("SSGA expansion search initialized");
 	}
@@ -102,6 +103,8 @@ public SSGABeliefExpansion(IPOMDP ip, int iterations) {
 
 		this.exploredBeliefs = new HashSet<DD>();
 		this.exploredBeliefs.addAll(this.initialBeliefs);
+
+		this.maxBeliefs = 200;
 	}
 
 	// ----------------------------------------------------------------------------------------
@@ -192,8 +195,27 @@ public void expand() {
 
 							}
 
-							if (minDist > 0.01)
-								this.exploredBeliefs.add(nextBelief);
+							if (minDist > 0.01) {
+
+								if (this.exploredBeliefs.size() < this.maxBeliefs) {
+									this.exploredBeliefs.add(nextBelief);
+								}
+
+								else {
+
+									LOGGER.warn("Max beliefs limit reached. stopping.");
+									break;
+//									DD beliefToRemove = null;
+//									
+//									for (DD belToRemove: this.exploredBeliefs) {
+//										beliefToRemove = belToRemove;
+//										break;
+//									}
+//									
+//									this.exploredBeliefs.remove(beliefToRemove);
+//									this.exploredBeliefs.add(nextBelief);
+								}
+							}
 						}
 
 						belief = nextBelief;

diff --git a/Protos/src/thinclab/belief/SparseFullBeliefExpansion.java b/Protos/src/thinclab/belief/SparseFullBeliefExpansion.java
@@ -32,6 +32,7 @@ public class SparseFullBeliefExpansion extends FullBeliefExpansion {
 	 */
 
 	private int nIters = 30;
+	private int maxBeliefs = 200;
 
 	private static final long serialVersionUID = 1571517162981801344L;
 
@@ -120,6 +121,20 @@ public void expandSingleStep() {
 				if (nextBelief == null) continue;
 
 				if (!this.exploredBeliefs.contains(nextBelief)) {
+
+					if (this.exploredBeliefs.size() > this.maxBeliefs) {
+
+						LOGGER.warn("Max beliefs limit reached, stopping");
+						break;
+//						DD beliefToRemove = null;
+//						for (DD belToRemove: this.exploredBeliefs) {
+//							beliefToRemove = belToRemove;
+//							break;
+//						}
+//						
+//						this.exploredBeliefs.remove(beliefToRemove);
+					}
+
 					this.exploredBeliefs.add(nextBelief);
 					newLeaves.add(nextBelief);
 				}

diff --git a/Protos/src/thinclab/decisionprocesses/DecisionProcess.java b/Protos/src/thinclab/decisionprocesses/DecisionProcess.java
@@ -60,9 +60,15 @@ public abstract class DecisionProcess implements Serializable {
 	public abstract void step(DD belief, String action, String[] obs) throws Exception;
 	public abstract void setTi(String action, DD[] Ti);
 	public abstract int getNumVars();
+
 	public abstract double evaluatePolicy(
 			DD[] alphaVectors, int[] policy, int trials, int evalDepth, boolean verbose);
 
+	public abstract double evaluateDefaultPolicy(
+			String defaultAction, int trials, int evalDepth, boolean verbose);
+
+	public abstract double evaluateRandomPolicy(int trials, int evalDepth, boolean verbose);
+
 	// ---------------------------------------------------------------------------------
 
 	public static String getActionFromPolicy(