Changeset 12815 for branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.MonteCarloTreeSearch

Timestamp:

07/30/15 16:37:14 (9 years ago)

Author:

aballeit

Message:

#2283 added SelectionIndicator for MCTS and problems SantaFeAnt, SymbolicRegression10, RoyalSymbol; updated SantaFeAnt grammar

Location:

branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.MonteCarloTreeSearch

Files:

• MonteCarloTreeSearch.cs (modified) (8 diffs)
• MonteCarloTreeSearch_PruneLeaves.cs (modified) (6 diffs)

branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.MonteCarloTreeSearch/MonteCarloTreeSearch.cs

@@ -1 +1 @@
 using System;
 using System.Collections.Generic;
+using System.Diagnostics;
 using System.Drawing;
 using System.Linq;
@@ -21 +22 @@
     {
         protected readonly int maxLen;
-        protected readonly IProblem problem;
+        protected readonly ISymbolicExpressionTreeProblem problem;
         protected readonly IGrammar grammar;
         protected readonly Random random;
@@ -29 +30 @@
         protected bool isPaused = false;
         protected object pauseLock = new object();
-
-        public MonteCarloTreeSearch(IProblem problem, int maxLen, Random random, IBanditPolicy behaviourPolicy,
+        protected int goodSelections;
+
+        public MonteCarloTreeSearch(ISymbolicExpressionTreeProblem problem, int maxLen, Random random, IBanditPolicy behaviourPolicy,
             ISimulation simulationPolicy)
         {
@@ -39 +41 @@
             this.behaviourPolicy = behaviourPolicy;
             this.simulation = simulationPolicy;
+
+            this.problem.GenerateProblemSolutions(maxLen);
+        }
+
+        public event Action<int, int> NodeSelected;
+
+        protected virtual void OnNodeSelectedChanged(int value, int selections)
+        {
+            RaiseNodeSelectedChanged(value, selections);
+        }
+
+        private void RaiseNodeSelectedChanged(int value, int selections)
+        {
+            var handler = NodeSelected;
+            if (handler != null) handler(value, selections);
         }
 
@@ -48 +65 @@
         }
 
+        public int GoodSelections { get { return this.goodSelections; } }
+
+        protected void CheckSelection(TreeNode node, int selections)
+        {
+            if (problem.IsParentOfProblemSolution(node.phrase, maxLen))
+            {
+                goodSelections++;
+            }
+
+            OnNodeSelectedChanged(goodSelections, selections);
+        }
+
         public override void Run(int maxIterations)
         {
             Reset();
+            int selections = 0;
             for (int i = 0; !StopRequested && i < maxIterations; i++)
             {
@@ -67 +97 @@
                         currentNode.GetChildActionInfos());
                     currentNode = currentNode.children[currentActionIndex];
+                    selections++;
+                    CheckSelection(currentNode, selections);
                 }
 
@@ -77 +109 @@
                         currentNode.children[behaviourPolicy.SelectAction(random, currentNode.GetChildActionInfos())
                             ];
+                    selections++;
+                    CheckSelection(currentNode, selections);
                 }
                 if (currentNode.phrase.Length <= maxLen)
@@ -122 +156 @@
         protected void Reset()
         {
+            goodSelections = 0;
             StopRequested = false;
             bestQuality = 0.0;

branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.MonteCarloTreeSearch/MonteCarloTreeSearch_PruneLeaves.cs

@@ -1 @@
-using HeuristicLab.Algorithms.Bandits;
+using System.Diagnostics;
+using HeuristicLab.Algorithms.Bandits;
 using HeuristicLab.Algorithms.MonteCarloTreeSearch.Base;
 using HeuristicLab.Algorithms.MonteCarloTreeSearch.Simulation;
@@ -11 +12 @@
     public class MonteCarloTreeSearch_PruneLeaves : MonteCarloTreeSearch
     {
-        public MonteCarloTreeSearch_PruneLeaves(IProblem problem, int maxLen, Random random, IBanditPolicy behaviourPolicy,
-            ISimulation simulationPolicy):base(problem,maxLen,random,behaviourPolicy,simulationPolicy)
+        public MonteCarloTreeSearch_PruneLeaves(ISymbolicExpressionTreeProblem problem, int maxLen, Random random, IBanditPolicy behaviourPolicy,
+            ISimulation simulationPolicy)
+            : base(problem, maxLen, random, behaviourPolicy, simulationPolicy)
         {
         }
@@ -19 +21 @@
         {
             Reset();
+            int selections = 0;
             for (int i = 0; !StopRequested && i < maxIterations; i++)
             {
@@ -35 +38 @@
                         currentNode.GetChildActionInfos());
                     currentNode = currentNode.children[currentActionIndex];
+                    selections++;
+                    CheckSelection(currentNode, selections);
                 }
 
@@ -54 +59 @@
                             currentNode.children[behaviourPolicy.SelectAction(random, currentNode.GetChildActionInfos())
                                 ];
+                        selections++;
+                        CheckSelection(currentNode, selections);
                     }
                     else
@@ -69 +76 @@
 
                     Propagate(currentNode, quality);
-
                 }
             }