Changeset 14251 for branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic

Timestamp:

08/10/16 20:10:25 (8 years ago)

Author:

gkronber

Message:

#2650:

• extended non-linear regression to work with factors
• fixed bugs in constants optimizer and tree interpreter
• improved simplification of factor variables
• added support for factors to ERC view
• added support for factors to solution comparison view
• activated view for all factors

Location:

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4

Files:

• Formatters/InfixExpressionFormatter.cs (modified) (1 diff)
• HeuristicLab.Problems.DataAnalysis.Symbolic-3.4.csproj (modified) (1 diff)
• Importer/InfixExpressionParser.cs (modified) (7 diffs)
• Interfaces/IVariableSymbol.cs (modified) (1 diff)
• Interfaces/IVariableTreeNode.cs (added)
• Interpreter/SymbolicDataAnalysisExpressionTreeInterpreter.cs (modified) (1 diff)
• SymbolicDataAnalysisExpressionTreeSimplifier.cs (modified) (47 diffs)
• Symbols/FactorVariableTreeNode.cs (modified) (1 diff)

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Formatters/InfixExpressionFormatter.cs

@@ -116 +116 @@
             strBuilder.Append(")");
           }
+        } else if (node.Symbol is FactorVariable) {
+          var factorNode = node as FactorVariableTreeNode;
+          if (factorNode.VariableName.Contains("'")) {
+            strBuilder.AppendFormat("\"{0}\"", factorNode.VariableName);
+          } else {
+            strBuilder.AppendFormat("'{0}'", factorNode.VariableName);
+          }
+        } else if (node.Symbol is BinaryFactorVariable) {
+          var factorNode = node as BinaryFactorVariableTreeNode;
+          if (!factorNode.Weight.IsAlmost(1.0)) {
+            strBuilder.Append("(");
+            strBuilder.AppendFormat(CultureInfo.InvariantCulture, "{0}", factorNode.Weight);
+            strBuilder.Append("*");
+          }
+          if (factorNode.VariableName.Contains("'")) {
+            strBuilder.AppendFormat("\"{0}={1}\"", factorNode.VariableName, factorNode.VariableValue);
+          } else {
+            strBuilder.AppendFormat("'{0}={1}'", factorNode.VariableName, factorNode.VariableValue);
+          }
+          if (!factorNode.Weight.IsAlmost(1.0)) {
+            strBuilder.Append(")");
+          }
+
         } else if (node.Symbol is Constant) {
           var constNode = node as ConstantTreeNode;

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/HeuristicLab.Problems.DataAnalysis.Symbolic-3.4.csproj

@@ -140 +140 @@
     <Compile Include="Importer\Token.cs" />
     <Compile Include="Interfaces\IModelBacktransformator.cs" />
+    <Compile Include="Interfaces\IVariableTreeNode.cs" />
     <Compile Include="Interfaces\IVariableSymbol.cs" />
     <Compile Include="Interpreter\SymbolicDataAnalysisExpressionCompiledTreeInterpreter.cs" />

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Importer/InfixExpressionParser.cs

@@ -32 +32 @@
   /// Parses mathematical expressions in infix form. E.g. x1 * (3.0 * x2 + x3)
   /// Identifier format (functions or variables): '_' | letter { '_' | letter | digit }
-  /// Variables names can be set under quotes "" or '' because variable names might contain spaces. 
+  /// Variables names and variable values can be set under quotes "" or '' because variable names might contain spaces. 
+  ///   Variable = ident | " ident " | ' ident ' 
   /// It is also possible to use functions e.g. log("x1") or real-valued constants e.g. 3.1415 . 
   /// Variable names are case sensitive. Function names are not case sensitive.
+  /// 
+  /// 
+  /// S       = Expr EOF
+  /// Expr    = ['-' | '+'] Term { '+' Term | '-' Term }
+  /// Term    = Fact { '*' Fact | '/' Fact }
+  /// Fact    = '(' Expr ')' | funcId '(' Expr ')' | VarExpr | number
+  /// VarExpr = varId [ '=' varVal]
+  /// varId   =  ident | ' ident ' | " ident "
+  /// varVal  =  ident | ' ident ' | " ident "
+  /// ident   =  '_' | letter { '_' | letter | digit }
   /// </summary>
   public sealed class InfixExpressionParser {
-    private enum TokenType { Operator, Identifier, Number, LeftPar, RightPar, End, NA };
+    private enum TokenType { Operator, Identifier, Number, LeftPar, RightPar, Eq, End, NA };
     private class Token {
       internal double doubleVal;
@@ -64 +75 @@
     private Constant constant = new Constant();
     private Variable variable = new Variable();
+    private BinaryFactorVariable binaryFactorVar = new BinaryFactorVariable();
 
     private ProgramRootSymbol programRootSymbol = new ProgramRootSymbol();
@@ -143 +155 @@
           pos++;
           while (pos < str.Length && !char.IsWhiteSpace(str[pos])
-            && (str[pos] != '+' || str[pos-1] == 'e' || str[pos-1] == 'E')     // continue reading exponents
+            && (str[pos] != '+' || str[pos - 1] == 'e' || str[pos - 1] == 'E')     // continue reading exponents
             && (str[pos] != '-' || str[pos - 1] == 'e' || str[pos - 1] == 'E')
-            && str[pos] != '*'           
+            && str[pos] != '*'
             && str[pos] != '/'
             && str[pos] != ')') {
@@ -211 +223 @@
           pos++;
           yield return new Token { TokenType = TokenType.RightPar, strVal = ")" };
-        }
-      }
-    }
-
-    // S = Expr EOF
-    // Expr = ['-' | '+'] Term { '+' Term | '-' Term }
-    // Term = Fact { '*' Fact | '/' Fact }
-    // Fact = '(' Expr ')' | funcId '(' Expr ')' | varId | number
+        } else if (str[pos] == '=') {
+          pos++;
+          yield return new Token { TokenType = TokenType.Eq, strVal = "=" };
+        }
+      }
+    }
+
     private ISymbolicExpressionTreeNode ParseS(Queue<Token> tokens) {
       var expr = ParseExpr(tokens);
@@ -326 +337 @@
     }
 
-    // Fact = '(' Expr ')' | funcId '(' Expr ')' | varId | number
+    // Fact = '(' Expr ')' | funcId '(' Expr ')' | varId [ = valId ] | number
     private ISymbolicExpressionTreeNode ParseFact(Queue<Token> tokens) {
       var next = tokens.Peek();
@@ -355 +366 @@
         } else {
           // variable
-          var varNode = (VariableTreeNode)variable.CreateTreeNode();
-          varNode.Weight = 1.0;
-          varNode.VariableName = idTok.strVal;
-          return varNode;
+          if (tokens.Peek().TokenType == TokenType.Eq) {
+            // binary factor
+            tokens.Dequeue(); // skip Eq
+            var valTok = tokens.Dequeue();
+            if (valTok.TokenType != TokenType.Identifier) throw new ArgumentException("expected identifier");
+            var binFactorNode = (BinaryFactorVariableTreeNode)binaryFactorVar.CreateTreeNode();
+            binFactorNode.Weight = 1.0;
+            binFactorNode.VariableName = idTok.strVal;
+            binFactorNode.VariableValue = valTok.strVal;
+            return binFactorNode;
+          } else {
+            // variable
+            var varNode = (VariableTreeNode)variable.CreateTreeNode();
+            varNode.Weight = 1.0;
+            varNode.VariableName = idTok.strVal;
+            return varNode;
+          }
         }
       } else if (next.TokenType == TokenType.Number) {
@@ -369 +393 @@
       }
     }
+
   }
 }

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interfaces/IVariableSymbol.cs

@@ -24 +24 @@
 
   /// <summary>
-  /// Any symbol that references a variable
+  /// Any symbol that references a variable (includes terminal as well as non-terminal symbols)
   /// </summary>
   public interface IVariableSymbol : ISymbol {
   }
-
-  /// <summary>
-  /// Any tree node that references a variable 
-  /// </summary>
-  public interface IVariableTreeNode : ISymbolicExpressionTreeNode {
-    string VariableName { get; }
-  }
 }

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interpreter/SymbolicDataAnalysisExpressionTreeInterpreter.cs

@@ -145 +145 @@
           instr.data = dataset.GetReadOnlyDoubleValues(variableTreeNode.VariableName);
         } else if (instr.opCode == OpCodes.FactorVariable) {
+          var factorTreeNode = instr.dynamicNode as FactorVariableTreeNode;
+          instr.data = dataset.GetReadOnlyStringValues(factorTreeNode.VariableName);
+        } else if (instr.opCode == OpCodes.BinaryFactorVariable) {
           var factorTreeNode = instr.dynamicNode as BinaryFactorVariableTreeNode;
           instr.data = dataset.GetReadOnlyStringValues(factorTreeNode.VariableName);

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/SymbolicDataAnalysisExpressionTreeSimplifier.cs

@@ -1 +1 @@
 #region License Information
+
 /* HeuristicLab
  * Copyright (C) 2002-2016 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
@@ -18 +19 @@
  * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
  */
+
 #endregion
 
@@ -24 +26 @@
 using System.Linq;
 using HeuristicLab.Common;
+using HeuristicLab.Core;
 using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
 
@@ -66 +69 @@
 
     // the argumentTrees list contains already expanded trees used as arguments for invocations
-    private ISymbolicExpressionTreeNode MacroExpand(ISymbolicExpressionTreeNode root, ISymbolicExpressionTreeNode node, IList<ISymbolicExpressionTreeNode> argumentTrees) {
+    private ISymbolicExpressionTreeNode MacroExpand(ISymbolicExpressionTreeNode root, ISymbolicExpressionTreeNode node,
+      IList<ISymbolicExpressionTreeNode> argumentTrees) {
       List<ISymbolicExpressionTreeNode> subtrees = new List<ISymbolicExpressionTreeNode>(node.Subtrees);
       while (node.SubtreeCount > 0) node.RemoveSubtree(0);
@@ -98 +102 @@
     }
 
-
     #region symbol predicates
+
     // arithmetic
     private bool IsDivision(ISymbolicExpressionTreeNode node) {
@@ -120 +124 @@
       return node.Symbol is Average;
     }
+
     // exponential
     private bool IsLog(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Logarithm;
     }
+
     private bool IsExp(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Exponential;
     }
+
     private bool IsRoot(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Root;
     }
+
     private bool IsSquare(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Square;
     }
+
     private bool IsSquareRoot(ISymbolicExpressionTreeNode node) {
       return node.Symbol is SquareRoot;
     }
+
     private bool IsPower(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Power;
     }
+
     // trigonometric
     private bool IsSine(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Sine;
     }
+
     private bool IsCosine(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Cosine;
     }
+
     private bool IsTangent(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Tangent;
     }
+
     // boolean
     private bool IsIfThenElse(ISymbolicExpressionTreeNode node) {
       return node.Symbol is IfThenElse;
     }
+
     private bool IsAnd(ISymbolicExpressionTreeNode node) {
       return node.Symbol is And;
     }
+
     private bool IsOr(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Or;
     }
+
     private bool IsNot(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Not;
     }
+
     // comparison
     private bool IsGreaterThan(ISymbolicExpressionTreeNode node) {
       return node.Symbol is GreaterThan;
     }
+
     private bool IsLessThan(ISymbolicExpressionTreeNode node) {
       return node.Symbol is LessThan;
@@ -182 +201 @@
       return node.Symbol is Variable;
     }
+
     private bool IsVariableBase(ISymbolicExpressionTreeNode node) {
       return node is VariableTreeNodeBase;
     }
+
+    private bool IsFactor(ISymbolicExpressionTreeNode node) {
+      return node is FactorVariableTreeNode;
+    }
+
     private bool IsConstant(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Constant;
@@ -193 +218 @@
       return node.Symbol is TimeLag;
     }
+
     private bool IsIntegral(ISymbolicExpressionTreeNode node) {
       return node.Symbol is Integral;
@@ -256 +282 @@
     }
 
-
     #region specific simplification routines
+
     private ISymbolicExpressionTreeNode SimplifyAny(ISymbolicExpressionTreeNode original) {
       // can't simplify this function but simplify all subtrees 
@@ -305 +331 @@
         var remaining = original.Subtrees.Skip(2);
         return
-          MakeProduct(GetSimplifiedTree(first), Invert(remaining.Aggregate(GetSimplifiedTree(second), (a, b) => MakeProduct(a, GetSimplifiedTree(b)))));
+          MakeProduct(GetSimplifiedTree(first),
+            Invert(remaining.Aggregate(GetSimplifiedTree(second), (a, b) => MakeProduct(a, GetSimplifiedTree(b)))));
       }
     }
@@ -346 +373 @@
       return MakeNot(GetSimplifiedTree(original.GetSubtree(0)));
     }
+
     private ISymbolicExpressionTreeNode SimplifyOr(ISymbolicExpressionTreeNode original) {
       return original.Subtrees
@@ -351 +379 @@
         .Aggregate(MakeOr);
     }
+
     private ISymbolicExpressionTreeNode SimplifyAnd(ISymbolicExpressionTreeNode original) {
       return original.Subtrees
@@ -356 +385 @@
         .Aggregate(MakeAnd);
     }
+
     private ISymbolicExpressionTreeNode SimplifyLessThan(ISymbolicExpressionTreeNode original) {
       return MakeLessThan(GetSimplifiedTree(original.GetSubtree(0)), GetSimplifiedTree(original.GetSubtree(1)));
     }
+
     private ISymbolicExpressionTreeNode SimplifyGreaterThan(ISymbolicExpressionTreeNode original) {
       return MakeGreaterThan(GetSimplifiedTree(original.GetSubtree(0)), GetSimplifiedTree(original.GetSubtree(1)));
     }
+
     private ISymbolicExpressionTreeNode SimplifyIfThenElse(ISymbolicExpressionTreeNode original) {
-      return MakeIfThenElse(GetSimplifiedTree(original.GetSubtree(0)), GetSimplifiedTree(original.GetSubtree(1)), GetSimplifiedTree(original.GetSubtree(2)));
-    }
+      return MakeIfThenElse(GetSimplifiedTree(original.GetSubtree(0)), GetSimplifiedTree(original.GetSubtree(1)),
+        GetSimplifiedTree(original.GetSubtree(2)));
+    }
+
     private ISymbolicExpressionTreeNode SimplifyTangent(ISymbolicExpressionTreeNode original) {
       return MakeTangent(GetSimplifiedTree(original.GetSubtree(0)));
     }
+
     private ISymbolicExpressionTreeNode SimplifyCosine(ISymbolicExpressionTreeNode original) {
       return MakeCosine(GetSimplifiedTree(original.GetSubtree(0)));
     }
+
     private ISymbolicExpressionTreeNode SimplifySine(ISymbolicExpressionTreeNode original) {
       return MakeSine(GetSimplifiedTree(original.GetSubtree(0)));
     }
+
     private ISymbolicExpressionTreeNode SimplifyExp(ISymbolicExpressionTreeNode original) {
       return MakeExp(GetSimplifiedTree(original.GetSubtree(0)));
     }
+
     private ISymbolicExpressionTreeNode SimplifySquare(ISymbolicExpressionTreeNode original) {
       return MakeSquare(GetSimplifiedTree(original.GetSubtree(0)));
     }
+
     private ISymbolicExpressionTreeNode SimplifySquareRoot(ISymbolicExpressionTreeNode original) {
       return MakeSquareRoot(GetSimplifiedTree(original.GetSubtree(0)));
@@ -387 +426 @@
       return MakeLog(GetSimplifiedTree(original.GetSubtree(0)));
     }
+
     private ISymbolicExpressionTreeNode SimplifyRoot(ISymbolicExpressionTreeNode original) {
       return MakeRoot(GetSimplifiedTree(original.GetSubtree(0)), GetSimplifiedTree(original.GetSubtree(1)));
@@ -394 +434 @@
       return MakePower(GetSimplifiedTree(original.GetSubtree(0)), GetSimplifiedTree(original.GetSubtree(1)));
     }
+
     private ISymbolicExpressionTreeNode SimplifyTimeLag(ISymbolicExpressionTreeNode original) {
       var laggedTreeNode = original as ILaggedTreeNode;
@@ -403 +444 @@
       }
     }
+
     private ISymbolicExpressionTreeNode SimplifyIntegral(ISymbolicExpressionTreeNode original) {
       var laggedTreeNode = original as ILaggedTreeNode;
@@ -416 +458 @@
 
     #region low level tree restructuring
+
     private ISymbolicExpressionTreeNode MakeTimeLag(ISymbolicExpressionTreeNode subtree, int lag) {
       if (lag == 0) return subtree;
@@ -446 +489 @@
       } else if (!IsBoolean(t)) {
         var gtNode = gtSymbol.CreateTreeNode();
-        gtNode.AddSubtree(t); gtNode.AddSubtree(MakeConstant(0.0));
+        gtNode.AddSubtree(t);
+        gtNode.AddSubtree(MakeConstant(0.0));
         var notNode = notSymbol.CreateTreeNode();
         notNode.AddSubtree(gtNode);
@@ -486 +530 @@
       }
     }
+
     private ISymbolicExpressionTreeNode MakeAnd(ISymbolicExpressionTreeNode a, ISymbolicExpressionTreeNode b) {
       if (IsConstant(a) && IsConstant(b)) {
@@ -515 +560 @@
       }
     }
-    private ISymbolicExpressionTreeNode MakeLessThan(ISymbolicExpressionTreeNode leftSide, ISymbolicExpressionTreeNode rightSide) {
+
+    private ISymbolicExpressionTreeNode MakeLessThan(ISymbolicExpressionTreeNode leftSide,
+      ISymbolicExpressionTreeNode rightSide) {
       if (IsConstant(leftSide) && IsConstant(rightSide)) {
         var lsConst = leftSide as ConstantTreeNode;
@@ -528 +575 @@
       }
     }
-    private ISymbolicExpressionTreeNode MakeGreaterThan(ISymbolicExpressionTreeNode leftSide, ISymbolicExpressionTreeNode rightSide) {
+
+    private ISymbolicExpressionTreeNode MakeGreaterThan(ISymbolicExpressionTreeNode leftSide,
+      ISymbolicExpressionTreeNode rightSide) {
       if (IsConstant(leftSide) && IsConstant(rightSide)) {
         var lsConst = leftSide as ConstantTreeNode;
@@ -541 +590 @@
       }
     }
-    private ISymbolicExpressionTreeNode MakeIfThenElse(ISymbolicExpressionTreeNode condition, ISymbolicExpressionTreeNode trueBranch, ISymbolicExpressionTreeNode falseBranch) {
+
+    private ISymbolicExpressionTreeNode MakeIfThenElse(ISymbolicExpressionTreeNode condition,
+      ISymbolicExpressionTreeNode trueBranch, ISymbolicExpressionTreeNode falseBranch) {
       if (IsConstant(condition)) {
         var constT = condition as ConstantTreeNode;
@@ -552 +603 @@
         } else {
           var gtNode = gtSymbol.CreateTreeNode();
-          gtNode.AddSubtree(condition); gtNode.AddSubtree(MakeConstant(0.0));
+          gtNode.AddSubtree(condition);
+          gtNode.AddSubtree(MakeConstant(0.0));
           ifNode.AddSubtree(gtNode);
         }
@@ -571 +623 @@
       }
     }
+
     private ISymbolicExpressionTreeNode MakeTangent(ISymbolicExpressionTreeNode node) {
       if (IsConstant(node)) {
@@ -581 +634 @@
       }
     }
+
     private ISymbolicExpressionTreeNode MakeCosine(ISymbolicExpressionTreeNode node) {
       if (IsConstant(node)) {
@@ -591 +645 @@
       }
     }
+
     private ISymbolicExpressionTreeNode MakeExp(ISymbolicExpressionTreeNode node) {
       if (IsConstant(node)) {
         var constT = node as ConstantTreeNode;
         return MakeConstant(Math.Exp(constT.Value));
+      } else if (IsFactor(node)) {
+        var factNode = node as FactorVariableTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++) factNode.Weights[i] = Math.Exp(factNode.Weights[i]);
+        return factNode;
       } else if (IsLog(node)) {
         return node.GetSubtree(0);
@@ -612 +671 @@
         var constT = node as ConstantTreeNode;
         return MakeConstant(constT.Value * constT.Value);
+      } else if (IsFactor(node)) {
+        var factNode = node as FactorVariableTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++) factNode.Weights[i] = factNode.Weights[i] * factNode.Weights[i];
+        return factNode;
       } else if (IsSquareRoot(node)) {
         return node.GetSubtree(0);
@@ -620 +683 @@
       }
     }
+
     private ISymbolicExpressionTreeNode MakeSquareRoot(ISymbolicExpressionTreeNode node) {
       if (IsConstant(node)) {
         var constT = node as ConstantTreeNode;
         return MakeConstant(Math.Sqrt(constT.Value));
+      } else if (IsFactor(node)) {
+        var factNode = node as FactorVariableTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++) factNode.Weights[i] = Math.Sqrt(factNode.Weights[i]);
+        return factNode;
       } else if (IsSquare(node)) {
         return node.GetSubtree(0);
@@ -637 +705 @@
         var constT = node as ConstantTreeNode;
         return MakeConstant(Math.Log(constT.Value));
+      } else if (IsFactor(node)) {
+        var factNode = node as FactorVariableTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++) factNode.Weights[i] = Math.Log(factNode.Weights[i]);
+        return factNode;
       } else if (IsExp(node)) {
         return node.GetSubtree(0);
@@ -647 +719 @@
       }
     }
+
     private ISymbolicExpressionTreeNode MakeRoot(ISymbolicExpressionTreeNode a, ISymbolicExpressionTreeNode b) {
       if (IsConstant(a) && IsConstant(b)) {
@@ -652 +725 @@
         var constB = b as ConstantTreeNode;
         return MakeConstant(Math.Pow(constA.Value, 1.0 / Math.Round(constB.Value)));
+      } else if (IsFactor(a) && IsConstant(b)) {
+        var factNode = a as FactorVariableTreeNode;
+        var constNode = b as ConstantTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++)
+          factNode.Weights[i] = Math.Pow(factNode.Weights[i], 1.0 / Math.Round(constNode.Value));
+        return factNode;
+      } else if (IsConstant(a) && IsFactor(b)) {
+        var constNode = a as ConstantTreeNode;
+        var factNode = b as FactorVariableTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++)
+          factNode.Weights[i] = Math.Pow(constNode.Value, 1.0 / Math.Round(factNode.Weights[i]));
+        return factNode;
+      } else if (IsFactor(a) && IsFactor(b) && AreSameVariable(a, b)) {
+        var node0 = a as FactorVariableTreeNode;
+        var node1 = b as FactorVariableTreeNode;
+        for (int i = 0; i < node0.Weights.Length; i++)
+          node0.Weights[i] = Math.Pow(node0.Weights[i], 1.0 / Math.Round(node1.Weights[i]));
+        return node0;
       } else if (IsConstant(b)) {
         var constB = b as ConstantTreeNode;
@@ -679 +770 @@
       }
     }
+
     private ISymbolicExpressionTreeNode MakePower(ISymbolicExpressionTreeNode a, ISymbolicExpressionTreeNode b) {
       if (IsConstant(a) && IsConstant(b)) {
@@ -684 +776 @@
         var constB = b as ConstantTreeNode;
         return MakeConstant(Math.Pow(constA.Value, Math.Round(constB.Value)));
+      } else if (IsFactor(a) && IsConstant(b)) {
+        var factNode = a as FactorVariableTreeNode;
+        var constNode = b as ConstantTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++)
+          factNode.Weights[i] = Math.Pow(factNode.Weights[i], Math.Round(constNode.Value));
+        return factNode;
+      } else if (IsConstant(a) && IsFactor(b)) {
+        var constNode = a as ConstantTreeNode;
+        var factNode = b as FactorVariableTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++)
+          factNode.Weights[i] = Math.Pow(constNode.Value, Math.Round(factNode.Weights[i]));
+        return factNode;
+      } else if (IsFactor(a) && IsFactor(b) && AreSameVariable(a, b)) {
+        var node0 = a as FactorVariableTreeNode;
+        var node1 = b as FactorVariableTreeNode;
+        for (int i = 0; i < node0.Weights.Length; i++)
+          node0.Weights[i] = Math.Pow(node0.Weights[i], Math.Round(node1.Weights[i]));
+        return node0;
       } else if (IsConstant(b)) {
         var constB = b as ConstantTreeNode;
@@ -718 +828 @@
         // fold constants
         return MakeConstant(((ConstantTreeNode)a).Value / ((ConstantTreeNode)b).Value);
-      } if (IsConstant(a) && !((ConstantTreeNode)a).Value.IsAlmost(1.0)) {
+      } else if ((IsConstant(a) && !((ConstantTreeNode)a).Value.IsAlmost(1.0))) {
         return MakeFraction(MakeConstant(1.0), MakeProduct(b, Invert(a)));
       } else if (IsVariableBase(a) && IsConstant(b)) {
@@ -725 +835 @@
         ((VariableTreeNodeBase)a).Weight /= constB;
         return a;
+      } else if (IsFactor(a) && IsConstant(b)) {
+        var factNode = a as FactorVariableTreeNode;
+        var constNode = b as ConstantTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++) factNode.Weights[i] = factNode.Weights[i] / constNode.Value;
+        return factNode;
+      } else if (IsFactor(a) && IsFactor(b) && AreSameVariable(a, b)) {
+        var node0 = a as FactorVariableTreeNode;
+        var node1 = b as FactorVariableTreeNode;
+        for (int i = 0; i < node0.Weights.Length; i++) node0.Weights[i] = node0.Weights[i] / node1.Weights[i];
+        return node0;
+      } else if (IsFactor(a)) {
+        return MakeFraction(MakeConstant(1.0), MakeProduct(b, Invert(a)));
       } else if (IsVariableBase(a) && IsVariableBase(b) && AreSameVariable(a, b)) {
         // cancel variables
@@ -733 +855 @@
         return a.Subtrees
           .Select(x => GetSimplifiedTree(x))
-         .Select(x => MakeFraction(x, b))
-         .Aggregate((c, d) => MakeSum(c, d));
+          .Select(x => MakeFraction(x, b))
+          .Aggregate((c, d) => MakeSum(c, d));
       } else if (IsMultiplication(a) && IsConstant(b)) {
         return MakeProduct(a, Invert(b));
@@ -769 +891 @@
         // x + 0 => x
         return a;
+      } else if (IsFactor(a) && IsConstant(b)) {
+        var factNode = a as FactorVariableTreeNode;
+        var constNode = b as ConstantTreeNode;
+        for (int i = 0; i < factNode.Weights.Length; i++) factNode.Weights[i] = factNode.Weights[i] + constNode.Value;
+        return factNode;
+      } else if (IsFactor(a) && IsFactor(b) && AreSameVariable(a, b)) {
+        var node0 = a as FactorVariableTreeNode;
+        var node1 = b as FactorVariableTreeNode;
+        for (int i = 0; i < node0.Weights.Length; i++) node0.Weights[i] = node0.Weights[i] + node1.Weights[i];
+        return node0;
       } else if (IsAddition(a) && IsAddition(b)) {
         // merge additions
@@ -835 +967 @@
       var subtrees = new List<ISymbolicExpressionTreeNode>(sum.Subtrees);
       while (sum.Subtrees.Any()) sum.RemoveSubtree(0);
-      var groupedVarNodes = from node in subtrees.OfType<VariableTreeNodeBase>()
+      var groupedVarNodes = from node in subtrees.OfType<IVariableTreeNode>()
+                            where node.SubtreeCount == 0
+                            // only consider terminal nodes
                             let lag = (node is LaggedVariableTreeNode) ? ((LaggedVariableTreeNode)node).Lag : 0
-                            let cat = (node is BinaryFactorVariableTreeNode) ? ((BinaryFactorVariableTreeNode)node).VariableValue : string.Empty
-                            group node by node.VariableName + cat + lag into g
+                            let cat =
+                              (node is BinaryFactorVariableTreeNode) ? ((BinaryFactorVariableTreeNode)node).VariableValue : string.Empty
+                            group node by node.VariableName + cat + lag
+        into g
                             select g;
-      var unchangedSubtrees = subtrees.Where(t => !(t is VariableTreeNodeBase));
+      var constant = (from node in subtrees.OfType<ConstantTreeNode>()
+                      select node.Value).DefaultIfEmpty(0.0).Sum();
+      var unchangedSubtrees = subtrees.Where(t => t.SubtreeCount > 0 || !(t is IVariableTreeNode) && !(t is ConstantTreeNode));
 
       foreach (var variableNodeGroup in groupedVarNodes) {
-        var weightSum = variableNodeGroup.Select(t => t.Weight).Sum();
-        var representative = variableNodeGroup.First();
-        representative.Weight = weightSum;
-        sum.AddSubtree(representative);
+        var firstNode = variableNodeGroup.First();
+        if (firstNode is VariableTreeNodeBase) {
+          var representative = firstNode as VariableTreeNodeBase;
+          var weightSum = variableNodeGroup.Cast<VariableTreeNodeBase>().Select(t => t.Weight).Sum();
+          representative.Weight = weightSum;
+          sum.AddSubtree(representative);
+        } else if (firstNode is FactorVariableTreeNode) {
+          var representative = firstNode as FactorVariableTreeNode;
+          foreach (var node in variableNodeGroup.Skip(1).Cast<FactorVariableTreeNode>()) {
+            for (int j = 0; j < representative.Weights.Length; j++) {
+              representative.Weights[j] += node.Weights[j];
+            }
+          }
+          for (int j = 0; j < representative.Weights.Length; j++) {
+            representative.Weights[j] += constant;
+          }
+          sum.AddSubtree(representative);
+        }
       }
       foreach (var unchangedSubtree in unchangedSubtrees)
         sum.AddSubtree(unchangedSubtree);
+      if (!constant.IsAlmost(0.0)) {
+        sum.AddSubtree(MakeConstant(constant));
+      }
     }
 
@@ -861 +1016 @@
         // a * $ => $ * a
         return MakeProduct(b, a);
+      } else if (IsFactor(a) && IsFactor(b) && AreSameVariable(a, b)) {
+        var node0 = a as FactorVariableTreeNode;
+        var node1 = b as FactorVariableTreeNode;
+        for (int i = 0; i < node0.Weights.Length; i++) node0.Weights[i] = node0.Weights[i] * node1.Weights[i];
+        return node0;
+      } else if (IsFactor(a) && IsConstant(b)) {
+        var node0 = a as FactorVariableTreeNode;
+        var node1 = b as ConstantTreeNode;
+        for (int i = 0; i < node0.Weights.Length; i++) node0.Weights[i] = node0.Weights[i] * node1.Value;
+        return node0;
       } else if (IsConstant(b) && ((ConstantTreeNode)b).Value.IsAlmost(1.0)) {
         // $ * 1.0 => $
@@ -868 +1033 @@
         ((VariableTreeNodeBase)a).Weight *= ((ConstantTreeNode)b).Value;
         return a;
-      } else if (IsConstant(b) && IsAddition(a)) {
+      } else if (IsConstant(b) && IsAddition(a) ||
+          IsFactor(a) && IsAddition(b) ||
+          IsAddition(b) && IsFactor(a)) {
         // multiply constants into additions
         return a.Subtrees.Select(x => MakeProduct(x, b)).Aggregate((c, d) => MakeSum(c, d));
@@ -902 +1069 @@
       }
     }
+
     #endregion
 
-
     #region helper functions
+
     private bool ContainsVariableCondition(ISymbolicExpressionTreeNode node) {
       if (node.Symbol is VariableCondition) return true;
@@ -940 +1108 @@
       if (aLaggedVar != null && bLaggedVar != null) {
         return aLaggedVar.VariableName == bLaggedVar.VariableName &&
-          aLaggedVar.Lag == bLaggedVar.Lag;
+               aLaggedVar.Lag == bLaggedVar.Lag;
       }
       var aVar = a as VariableTreeNode;
@@ -947 +1115 @@
         return aVar.VariableName == bVar.VariableName;
       }
-      var aFactor = a as BinaryFactorVariableTreeNode;
-      var bFactor = b as BinaryFactorVariableTreeNode;
+      var aFactor = a as FactorVariableTreeNode;
+      var bFactor = b as FactorVariableTreeNode;
       if (aFactor != null && bFactor != null) {
-        return aFactor.VariableName == bFactor.VariableName &&
-          aFactor.VariableValue == bFactor.VariableValue;
-      }
-
+        return aFactor.VariableName == bFactor.VariableName;
+      }
+      var aBinFactor = a as BinaryFactorVariableTreeNode;
+      var bBinFactor = b as BinaryFactorVariableTreeNode;
+      if (aBinFactor != null && bBinFactor != null) {
+        return aBinFactor.VariableName == bBinFactor.VariableName &&
+               aBinFactor.VariableValue == bBinFactor.VariableValue;
+      }
       return false;
     }
@@ -961 +1133 @@
       var subtrees = new List<ISymbolicExpressionTreeNode>(prod.Subtrees);
       while (prod.Subtrees.Any()) prod.RemoveSubtree(0);
-      var groupedVarNodes = from node in subtrees.OfType<VariableTreeNodeBase>()
+      var groupedVarNodes = from node in subtrees.OfType<IVariableTreeNode>()
+                            where node.SubtreeCount == 0
                             let lag = (node is LaggedVariableTreeNode) ? ((LaggedVariableTreeNode)node).Lag : 0
-                            group node by node.VariableName + lag into g
+                            let cat =
+                              (node is BinaryFactorVariableTreeNode) ? ((BinaryFactorVariableTreeNode)node).VariableValue : string.Empty
+                            group node by node.VariableName + cat + lag
+        into g
                             orderby g.Count()
                             select g;
       var constantProduct = (from node in subtrees.OfType<VariableTreeNodeBase>()
                              select node.Weight)
-                            .Concat(from node in subtrees.OfType<ConstantTreeNode>()
-                                    select node.Value)
-                            .DefaultIfEmpty(1.0)
-                            .Aggregate((c1, c2) => c1 * c2);
+        .Concat(from node in subtrees.OfType<ConstantTreeNode>()
+                select node.Value)
+        .DefaultIfEmpty(1.0)
+        .Aggregate((c1, c2) => c1 * c2);
 
       var unchangedSubtrees = from tree in subtrees
-                              where !(tree is VariableTreeNodeBase)
-                              where !(tree is ConstantTreeNode)
+                              where tree.SubtreeCount > 0 || !(tree is IVariableTreeNode) && !(tree is ConstantTreeNode)
                               select tree;
 
       foreach (var variableNodeGroup in groupedVarNodes) {
-        var representative = variableNodeGroup.First();
-        representative.Weight = 1.0;
-        if (variableNodeGroup.Count() > 1) {
-          var poly = mulSymbol.CreateTreeNode();
-          for (int p = 0; p < variableNodeGroup.Count(); p++) {
-            poly.AddSubtree((ISymbolicExpressionTreeNode)representative.Clone());
+        var firstNode = variableNodeGroup.First();
+        if (firstNode is VariableTreeNodeBase) {
+          var representative = (VariableTreeNodeBase)firstNode;
+          representative.Weight = 1.0;
+          if (variableNodeGroup.Count() > 1) {
+            var poly = mulSymbol.CreateTreeNode();
+            for (int p = 0; p < variableNodeGroup.Count(); p++) {
+              poly.AddSubtree((ISymbolicExpressionTreeNode)representative.Clone());
+            }
+            prod.AddSubtree(poly);
+          } else {
+            prod.AddSubtree(representative);
           }
-          prod.AddSubtree(poly);
-        } else {
+        } else if (firstNode is FactorVariableTreeNode) {
+          var representative = (FactorVariableTreeNode)firstNode;
+          foreach (var node in variableNodeGroup.Skip(1).Cast<FactorVariableTreeNode>()) {
+            for (int j = 0; j < representative.Weights.Length; j++) {
+              representative.Weights[j] *= node.Weights[j];
+            }
+          }
+          for (int j = 0; j < representative.Weights.Length; j++) {
+            representative.Weights[j] *= constantProduct;
+          }
+          constantProduct = 1.0;
+          // if the product already contains a factor it is not necessary to multiply a constant below
           prod.AddSubtree(representative);
         }
@@ -1013 +1204 @@
         var variableTree = (VariableTreeNodeBase)x;
         variableTree.Weight *= -1.0;
+      } else if (IsFactor(x)) {
+        var factorNode = (FactorVariableTreeNode)x;
+        for (int i = 0; i < factorNode.Weights.Length; i++) factorNode.Weights[i] *= -1;
       } else if (IsAddition(x)) {
         // (x0 + x1 + .. + xn) * -1 => (-x0 + -x1 + .. + -xn)        
@@ -1041 +1235 @@
       if (IsConstant(x)) {
         return MakeConstant(1.0 / ((ConstantTreeNode)x).Value);
+      } else if (IsFactor(x)) {
+        var factorNode = (FactorVariableTreeNode)x;
+        for (int i = 0; i < factorNode.Weights.Length; i++) factorNode.Weights[i] = 1.0 / factorNode.Weights[i];
+        return factorNode;
       } else if (IsDivision(x)) {
         return MakeFraction(x.GetSubtree(1), x.GetSubtree(0));
@@ -1061 +1259 @@
       return tree;
     }
+
     #endregion
   }

branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Symbols/FactorVariableTreeNode.cs

@@ -114 +114 @@
 
     public override string ToString() {
-      return VariableName + "(factor)";
+      return VariableName + " (factor)";
     }
   }