Changeset 12298 for branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Problems.GrammaticalOptimization

Timestamp:

04/10/15 16:12:08 (9 years ago)

Author:

gkronber

Message:

#2283: experiments with q-learning

Location:

branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Problems.GrammaticalOptimization

Files:

• PartialExpressionInterpreter.cs (modified) (8 diffs)
• Problems/SymbolicRegressionPoly10Problem.cs (modified) (1 diff)
• SentenceSetStatistics.cs (modified) (1 diff)

branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Problems.GrammaticalOptimization/PartialExpressionInterpreter.cs

@@ -2 +2 @@
 using System.Collections.Generic;
 using System.Linq;
+using System.Security.Policy;
 using HeuristicLab.Common;
 
@@ -9 +10 @@
     private string sentence;
     private int syIdx;
+    private HashSet<double> intermediateValues = new HashSet<double>();
     private Stack<double> stack = new Stack<double>();
-    private Stack<char> opStack = new Stack<char>();
     // interprets sentences from L(G(Expr)):
     // Expr -> Term { ('+' | '-' | '^' ) Term } 
@@ -22 +23 @@
     // The constant symbols '0' .. '9' are treated as ERC indices
 
-    public Stack<double> Interpret(string sentence, double[] vars) {
+    public IEnumerable<double> Interpret(string sentence, double[] vars) {
       return Interpret(sentence, vars, emptyErc);
     }
 
-    public Stack<double> Interpret(string sentence, double[] vars, double[] erc) {
+    public IEnumerable<double> Interpret(string sentence, double[] vars, double[] erc) {
       InitLex(sentence);
-      stack.Clear(); opStack.Clear();
+      intermediateValues.Clear();
+      stack.Clear();
       Expr(vars, erc);
-      return new Stack<double>(stack);
+      return intermediateValues;
     }
 
@@ -58 +60 @@
         if (curSy == '+') {
           NewSy();
-          if (!Term(d, erc)) { stack.Push(-1.0); return false; }
+          if (!Term(d, erc)) { return false; }
           stack.Push(stack.Pop() + stack.Pop());
+          intermediateValues.Add(stack.Peek());
         } else if (curSy == '-') {
           NewSy();
-          if (!Term(d, erc)) { stack.Push(-2.0); return false; return false; }
+          if (!Term(d, erc)) {  return false; }
           stack.Push(-stack.Pop() + stack.Pop());
+          intermediateValues.Add(stack.Peek());
         } else {
           NewSy();
-          if (!Term(d, erc)) { stack.Push(-3.0); return false; }
+          if (!Term(d, erc)) { return false; }
           var e = stack.Pop();
           var r = stack.Pop();
           stack.Push(Not(r) * e + r * Not(e)); // xor = (!x AND y) OR (x AND !y)
+          intermediateValues.Add(stack.Peek());
         }
         curSy = CurSy();
@@ -82 +87 @@
         if (curSy == '*') {
           NewSy();
-          if (!Fact(d, erc)) { stack.Push(-4.0); return false; }
+          if (!Fact(d, erc)) { return false; }
           stack.Push(stack.Pop() * stack.Pop());
+          intermediateValues.Add(stack.Peek());
         } else {
           NewSy();
-          if (!Fact(d, erc)) { stack.Push(-5.0); return false; }
+          if (!Fact(d, erc)) { return false; }
           var nom = stack.Pop();
           var r = stack.Pop();
           if (HeuristicLab.Common.Extensions.IsAlmost(nom, 0.0)) nom = 1.0;
           stack.Push(r / nom);
+          intermediateValues.Add(stack.Peek());
         }
         curSy = CurSy();
@@ -100 +107 @@
       var curSy = CurSy();
       if (curSy == '!') {
-        NewSy();
-        if (!Expr(d, erc)) { stack.Push(-7.0); return false; }
-        stack.Push(Not(stack.Pop()));
+        //NewSy();
+        //if (!Expr(d, erc)) { stack.Push(-7.0); return false; }
+        //stack.Push(Not(stack.Pop()));
       } else if (curSy == '(') {
-        NewSy();
-        if (!Expr(d, erc)) { stack.Push(-8.0); return false; }
-        if (CurSy() != ')') throw new ArgumentException();
-        NewSy();
+        //NewSy();
+        //if (!Expr(d, erc)) { stack.Push(-8.0); return false; }
+        //if (CurSy() != ')') throw new ArgumentException();
+        //NewSy();
       } else if (curSy >= 'a' && curSy <= 'z') {
         int o = (byte)curSy - (byte)'a';
@@ -113 +120 @@
         if (o < 0 || o >= d.Length) throw new ArgumentException();
         stack.Push(d[o]);
+        intermediateValues.Add(stack.Peek());
         NewSy();
       } else if (curSy == '/') {
         // /-symbol is used in the expressionextender to represent inverse (1/x). 
         // this is necessary because we also use symbols 0..9 as indices for ERCs
-        NewSy();
-        if (!Fact(d, erc)) { stack.Push(-9.0); return false; }
-        stack.Push(1.0 / stack.Pop());
+        //NewSy();
+        //if (!Fact(d, erc)) { stack.Push(-9.0); return false; }
+        //stack.Push(1.0 / stack.Pop());
       } else if (curSy >= '0' && curSy <= '9') {
         int o = (byte)curSy - (byte)'0';
@@ -125 +133 @@
         if (o < 0 || o >= 10) throw new ArgumentException();
         stack.Push(erc[o]);
+        intermediateValues.Add(stack.Peek());
         NewSy();
       } else {

branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Problems.GrammaticalOptimization/Problems/SymbolicRegressionPoly10Problem.cs

@@ -156 +156 @@
     // splits the phrase into terms and creates (sparse) term-occurrance features
     public IEnumerable<Feature> GetFeatures(string phrase) {
-      // var canonicalTerms = new HashSet<string>();
-      // foreach (string t in phrase.Split('+')) {
-      //   canonicalTerms.Add(CanonicalTerm(t));
+      //if (phrase.EndsWith("E")) phrase = phrase.TrimEnd('*', '+', 'E');
+      //yield return new Feature("$$$", 1.0); // const
+      //var canonicalTerms = new HashSet<string>();
+      //foreach (string t in phrase.Split('+')) {
+      //  canonicalTerms.Add(CanonicalTerm(t));
+      //}
+      //return canonicalTerms.Select(entry => new Feature(entry, 1.0));
+      //.Concat(new Feature[] { new Feature(CanonicalRepresentation(phrase), 1.0) });
+
+
+      if (phrase.EndsWith("E")) phrase = phrase.TrimEnd('*', '+', 'E');
+      //var len = 5;
+      //var start = Math.Max(0, phrase.Length - len);
+      //var end = Math.Min(phrase.Length, start + len);
+      //string f = phrase.Substring(start, end - start);
+      //yield return new Feature(f, 1.0);
+      //
+
+      var terms = phrase.Split('+');
+      foreach (var t in terms.Distinct()) yield return new Feature(t, 1.0);
+
+      for (int i = 0; i < terms.Length; i++) {
+        for (int j = i + 1; j < terms.Length; j++) {
+          yield return new Feature(terms[i] + " " + terms[j], 1.0);
+        }
+      }
+
+      // var substrings = new HashSet<string>();
+      // for (int start = 0; start <= phrase.Length - 2; start += 2) {
+      //   var s = phrase.Substring(start, 3);
+      //   substrings.Add(s);
       // }
-      // return canonicalTerms.Select(entry => new Feature(entry, 1.0))
-      //   .Concat(new Feature[] { new Feature(CanonicalRepresentation(phrase), 1.0) });
-
-      return new Feature[] { new Feature(phrase, 1.0), };
-
-      // var partialInterpreter = new PartialExpressionInterpreter();
-      // var vars = new double[] { 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, };
-      // var s = partialInterpreter.Interpret(phrase, vars);
-      // //if (s.Any())
-      // //  return new Feature[] { new Feature(s.Pop().ToString(), 1.0), };
-      // //else
-      // //  return new Feature[] { new Feature("$", 1.0), };
-      // return new Feature[] { new Feature(string.Join(",", s), 1.0) };
+      // 
+      // var list = new List<string>(substrings);
+      // 
+      // for (int i = 0; i < list.Count; i++) {
+      //   yield return new Feature(list[i], 1.0);
+      //   //for (int j = i+1; j < list.Count; j++) {
+      //   //  yield return new Feature(list[i] + " " + list[j], 1.0);
+      //   //}
+      // }
+
+      //
+      // for (int len = 1; len <= phrase.Length; len += 2) {
+      //   var start = Math.Max(0, phrase.Length - len);
+      //   var end = Math.Min(phrase.Length, start + len);
+      //   string f = phrase.Substring(start, end - start);
+      //   yield return new Feature(f, 1.0);
+      // 
+      // }
+
+      //var partialInterpreter = new PartialExpressionInterpreter();
+      //var vars = new double[] { 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, };
+      //var s = partialInterpreter.Interpret(phrase, vars);
+      ////if (s.Any())
+      ////  return new Feature[] { new Feature(s.Pop().ToString(), 1.0), };
+      ////else
+      ////  return new Feature[] { new Feature("$", 1.0), };
+      //return s.Select(f => new Feature(f.ToString(), 1.0));
     }
 

branches/HeuristicLab.Problems.GrammaticalOptimization-gkr/HeuristicLab.Problems.GrammaticalOptimization/SentenceSetStatistics.cs

@@ -50 +50 @@
     public override string ToString() {
       return
-        string.Format("Sentences: {0,10} avg.-quality {1,7:F5} best {2,7:F5} {3,2} {4,10} {5,30} first {6,7:F5} {7,20} last {8,7:F5} {9,20}",
+        string.Format("Sentences: {0,10} avg.-quality {1,7:F5} best {2,7:F5} {3,2} {4,10} {5,30} last {6,7:F5} {7,20}",
       NumberOfSentences, AverageQuality,
       BestSentenceQuality, DoubleExtensions.IsAlmost(BestSentenceQuality, bestKnownQuality) ? 1.0 : 0.0,
       BestSentenceIndex, TrimToSize(BestSentence, 30),
-      FirstSentenceQuality, TrimToSize(FirstSentence, 20),
       LastSentenceQuality, TrimToSize(LastSentence, 20)
+      //LastSentenceQuality, TrimToSize(LastSentence, 20)
      );
     }