Changeset 14345 for trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic

Timestamp:

10/21/16 17:59:36 (8 years ago)

Author:

gkronber

Message:

#2690: implemented methods to generate symbolic expression tree solutions for decision tree models (random forest and gradient boosted) as well as views which make it possible to inspect each of the individual trees in a GBT and RF solution

Location:

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4

Files:

• Interpreter/SymbolicDataAnalysisExpressionCompiledTreeInterpreter.cs (modified) (1 diff)
• Interpreter/SymbolicDataAnalysisExpressionTreeInterpreter.cs (modified) (1 diff)
• Interpreter/SymbolicDataAnalysisExpressionTreeLinearInterpreter.cs (modified) (3 diffs)
• Symbols/VariableCondition.cs (modified) (1 diff)
• Symbols/VariableConditionTreeNode.cs (modified) (1 diff)

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interpreter/SymbolicDataAnalysisExpressionCompiledTreeInterpreter.cs

@@ -614 +614 @@
         case OpCodes.VariableCondition: {
             var variableConditionTreeNode = (VariableConditionTreeNode)node;
+            if (variableConditionTreeNode.Symbol.IgnoreSlope) throw new NotSupportedException("Strict variable conditionals are not supported");
             var variableName = variableConditionTreeNode.VariableName;
             var indexExpr = Expression.Constant(variableIndices[variableName]);

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interpreter/SymbolicDataAnalysisExpressionTreeInterpreter.cs

@@ -471 +471 @@
             if (row < 0 || row >= dataset.Rows) return double.NaN;
             var variableConditionTreeNode = (VariableConditionTreeNode)currentInstr.dynamicNode;
-            double variableValue = ((IList<double>)currentInstr.data)[row];
-            double x = variableValue - variableConditionTreeNode.Threshold;
-            double p = 1 / (1 + Math.Exp(-variableConditionTreeNode.Slope * x));
-
-            double trueBranch = Evaluate(dataset, ref row, state);
-            double falseBranch = Evaluate(dataset, ref row, state);
-
-            return trueBranch * p + falseBranch * (1 - p);
+            if (!variableConditionTreeNode.Symbol.IgnoreSlope) {
+              double variableValue = ((IList<double>)currentInstr.data)[row];
+              double x = variableValue - variableConditionTreeNode.Threshold;
+              double p = 1 / (1 + Math.Exp(-variableConditionTreeNode.Slope * x));
+
+              double trueBranch = Evaluate(dataset, ref row, state);
+              double falseBranch = Evaluate(dataset, ref row, state);
+
+              return trueBranch * p + falseBranch * (1 - p);
+            } else {
+              // strict threshold
+              double variableValue = ((IList<double>)currentInstr.data)[row];
+              if (variableValue <= variableConditionTreeNode.Threshold) {
+                var left = Evaluate(dataset, ref row, state);
+                state.SkipInstructions();
+                return left;
+              } else {
+                state.SkipInstructions();
+                return Evaluate(dataset, ref row, state);
+              }
+            }
           }
         default:

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Interpreter/SymbolicDataAnalysisExpressionTreeLinearInterpreter.cs

@@ -126 +126 @@
     private readonly object syncRoot = new object();
     public IEnumerable<double> GetSymbolicExpressionTreeValues(ISymbolicExpressionTree tree, IDataset dataset, IEnumerable<int> rows) {
+      if (!rows.Any()) return Enumerable.Empty<double>();
       if (CheckExpressionsWithIntervalArithmetic)
         throw new NotSupportedException("Interval arithmetic is not yet supported in the symbolic data analysis interpreter.");
@@ -159 +160 @@
           if (row < 0 || row >= dataset.Rows) instr.value = double.NaN;
           var variableConditionTreeNode = (VariableConditionTreeNode)instr.dynamicNode;
-          double variableValue = ((IList<double>)instr.data)[row];
-          double x = variableValue - variableConditionTreeNode.Threshold;
-          double p = 1 / (1 + Math.Exp(-variableConditionTreeNode.Slope * x));
-
-          double trueBranch = code[instr.childIndex].value;
-          double falseBranch = code[instr.childIndex + 1].value;
-
-          instr.value = trueBranch * p + falseBranch * (1 - p);
+          if (!variableConditionTreeNode.Symbol.IgnoreSlope) {
+            double variableValue = ((IList<double>)instr.data)[row];
+            double x = variableValue - variableConditionTreeNode.Threshold;
+            double p = 1 / (1 + Math.Exp(-variableConditionTreeNode.Slope * x));
+
+            double trueBranch = code[instr.childIndex].value;
+            double falseBranch = code[instr.childIndex + 1].value;
+
+            instr.value = trueBranch * p + falseBranch * (1 - p);
+          } else {
+            double variableValue = ((IList<double>)instr.data)[row];
+            if (variableValue <= variableConditionTreeNode.Threshold) {
+              instr.value = code[instr.childIndex].value;
+            } else {
+              instr.value = code[instr.childIndex + 1].value;
+            }
+          }
         } else if (instr.opCode == OpCodes.Add) {
           double s = code[instr.childIndex].value;
@@ -411 +421 @@
               for (int j = 1; j != seq.Length; ++j)
                 seq[j].skip = true;
-            }
-            break;
+              break;
+            }
         }
         #endregion

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Symbols/VariableCondition.cs

@@ -149 +149 @@
         }
       }
+    }
+
+    /// <summary>
+    /// Flag to indicate if the interpreter should ignore the slope parameter (introduced for representation of expression trees)
+    /// </summary>
+    [Storable]
+    private bool ignoreSlope;
+    public bool IgnoreSlope {
+      get { return ignoreSlope; }
+      set { ignoreSlope = value; }
     }
 

trunk/sources/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/Symbols/VariableConditionTreeNode.cs

@@ -96 +96 @@
 
     public override string ToString() {
-      if (slope.IsAlmost(0.0))
+      if (slope.IsAlmost(0.0) || Symbol.IgnoreSlope) {
+        return variableName + " < " + threshold.ToString("E4");
+      } else {
         return variableName + " > " + threshold.ToString("E4") + Environment.NewLine +
-          "slope: " + slope.ToString("E4");
-      else
-        return variableName + " > " + threshold.ToString("E4");
+               "slope: " + slope.ToString("E4");
+      }
     }
   }