Changeset 4475 for branches/DataAnalysis/HeuristicLab.Problems.DataAnalysis.MultiVariate.TimeSeriesPrognosis.Views/3.3

Timestamp:

09/23/10 13:49:30 (14 years ago)

Author:

gkronber

Message:

Fixed bugs in time series prognosis classes #1142.

File:

• branches/DataAnalysis/HeuristicLab.Problems.DataAnalysis.MultiVariate.TimeSeriesPrognosis.Views/3.3/InteractiveSymbolicTimeSeriesPrognosisSolutionSimplifierView.cs (modified) (3 diffs)

branches/DataAnalysis/HeuristicLab.Problems.DataAnalysis.MultiVariate.TimeSeriesPrognosis.Views/3.3/InteractiveSymbolicTimeSeriesPrognosisSolutionSimplifierView.cs

@@ -95 +95 @@
         int samplesEnd = Content.ProblemData.TrainingSamplesEnd.Value;
 
-        double[] alpha, beta;
         double quality;
         conditionVariableName = Content.ConditionalEvaluationVariable;
         targetVariables = Content.ProblemData.TargetVariables.CheckedItems.Select(x => x.Value.Value);
         int nTargetVariables = Content.ProblemData.TargetVariables.CheckedItems.Count();
-        lowerEstimationLimit = new DoubleArray(Enumerable.Repeat(double.NegativeInfinity, nTargetVariables).ToArray());
-        upperEstimationLimit = new DoubleArray(Enumerable.Repeat(double.PositiveInfinity, nTargetVariables).ToArray());
         interpreter = Content.Model.Interpreter;
         horizon = Content.Horizon;
         IEnumerable<int> rows = Enumerable.Range(samplesStart, samplesEnd - samplesStart);
-        SymbolicTimeSeriesPrognosisScaledNormalizedMseEvaluator.CalculateScalingParameters(simplifiedExpressionTree,
-          Content.ProblemData, interpreter,
-          conditionVariableName, rows,
-          out beta, out alpha);
-
-        quality = SymbolicTimeSeriesPrognosisScaledNormalizedMseEvaluator.Evaluate(simplifiedExpressionTree, Content.ProblemData, interpreter,
-          conditionVariableName, rows, horizon,
-          lowerEstimationLimit, upperEstimationLimit,
-          beta, alpha);
+        if (!string.IsNullOrEmpty(conditionVariableName)) {
+          rows = (from row in rows
+                  where !Content.ProblemData.Dataset[conditionVariableName, row].IsAlmost(0.0)
+                  select row).ToList();
+        }
+        lowerEstimationLimit = new DoubleArray(nTargetVariables);
+        upperEstimationLimit = new DoubleArray(nTargetVariables);
+        for (int i = 0; i < nTargetVariables; i++) {
+          lowerEstimationLimit[i] = Content.GetLowerEstimationLimit(i);
+          upperEstimationLimit[i] = Content.GetUpperEstimationLimit(i);
+        }
+        quality = SymbolicTimeSeriesPrognosisScaledNormalizedMseEvaluator.Calculate(simplifiedExpressionTree, Content.ProblemData, interpreter,
+          rows, horizon,
+          lowerEstimationLimit, upperEstimationLimit);
 
         this.CalculateReplacementNodes();
 
-        this.CalculateNodeImpacts(simplifiedExpressionTree, simplifiedExpressionTree.Root, quality);
+        this.CalculateNodeImpacts(simplifiedExpressionTree, simplifiedExpressionTree.Root, quality, rows);
         // show only interesing part
         this.treeChart.Tree = new SymbolicExpressionTree(simplifiedExpressionTree.Root.SubTrees[0]);
@@ -134 +136 @@
           while (start.SubTrees.Count > 0) start.RemoveSubTree(0);
           start.AddSubTree(node);
-          double constantTreeNodeValue = interpreter.GetSymbolicExpressionTreeValues(tree, Content.ProblemData.Dataset, targetVariables, trainingSamples, 1).Select(x => x[0]).Median();
+          // we only want a scalar replacement value for the single branch that should be evaluated
+          // so assume we only create an estimation for the first target variable
+          double constantTreeNodeValue = interpreter.GetSymbolicExpressionTreeValues(tree, Content.ProblemData.Dataset, targetVariables.Take(1) , trainingSamples, 1).Select(x => x[0]).Median();
           ConstantTreeNode constantTreeNode = MakeConstantTreeNode(constantTreeNodeValue);
           replacementNodes[node] = constantTreeNode;
@@ -141 +145 @@
     }
 
-    private void CalculateNodeImpacts(SymbolicExpressionTree tree, SymbolicExpressionTreeNode currentTreeNode, double originalQuality) {
+    private void CalculateNodeImpacts(SymbolicExpressionTree tree, SymbolicExpressionTreeNode currentTreeNode, double originalQuality, IEnumerable<int> rows) {
       foreach (SymbolicExpressionTreeNode childNode in currentTreeNode.SubTrees.ToList()) {
         if (!(childNode.Symbol is StartSymbol || childNode.Symbol is ProgramRootSymbol)) {
           SwitchNode(currentTreeNode, childNode, replacementNodes[childNode]);
-          int samplesStart = Content.ProblemData.TrainingSamplesStart.Value;
-          int samplesEnd = Content.ProblemData.TrainingSamplesEnd.Value;
-          double[] alpha;
-          double[] beta;
           int horizon = Content.Horizon;
-          IEnumerable<int> rows = Enumerable.Range(samplesStart, samplesEnd - samplesStart);          
-          SymbolicTimeSeriesPrognosisScaledNormalizedMseEvaluator.CalculateScalingParameters(tree,
-            Content.ProblemData, interpreter,
-            conditionVariableName, rows,
-            out beta, out alpha);
-
-          double newQuality = SymbolicTimeSeriesPrognosisScaledNormalizedMseEvaluator.Evaluate(tree, Content.ProblemData, interpreter,
-            conditionVariableName, rows, horizon,
-            lowerEstimationLimit, upperEstimationLimit,
-            beta, alpha);
+
+          double newQuality = SymbolicTimeSeriesPrognosisScaledNormalizedMseEvaluator.Calculate(tree, Content.ProblemData, interpreter,
+            rows, horizon,
+            lowerEstimationLimit, upperEstimationLimit);
 
           nodeImpacts[childNode] = newQuality / originalQuality;
           SwitchNode(currentTreeNode, replacementNodes[childNode], childNode);
         }
-        CalculateNodeImpacts(tree, childNode, originalQuality);
+        CalculateNodeImpacts(tree, childNode, originalQuality, rows);
       }
     }