Changeset 15974 for branches/2886_SymRegGrammarEnumeration/HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration/Analysis

Timestamp:

06/28/18 11:35:13 (6 years ago)

Author:

bburlacu

Message:

#2886: implement LRU cache for storing search nodes, introduce SortedSet for handling priorities, fix serialization and cloning

File:

• branches/2886_SymRegGrammarEnumeration/HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration/Analysis/RSquaredEvaluator.cs (modified) (5 diffs)

branches/2886_SymRegGrammarEnumeration/HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration/Analysis/RSquaredEvaluator.cs

@@ -16 +16 @@
   public class RSquaredEvaluator : Item, IGrammarEnumerationAnalyzer {
     public static readonly string BestTrainingQualityResultName = "Best R² (Training)";
+    public static readonly string BestTestQualityResultName = "Best R² (Test)";
     public static readonly string BestTrainingModelResultName = "Best model (Training)";
     public static readonly string BestTrainingSolutionResultName = "Best solution (Training)";
+    public static readonly string BestComplexityResultName = "Best solution complexity";
 
     private static readonly ISymbolicDataAnalysisExpressionTreeInterpreter expressionTreeLinearInterpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
@@ -42 +44 @@
       algorithm.Started -= OnStarted;
       algorithm.Stopped -= OnStopped;
-
       algorithm.DistinctSentenceGenerated -= AlgorithmOnDistinctSentenceGenerated;
     }
@@ -57 +58 @@
     }
 
-    private void OnStopped(object sender, EventArgs eventArgs) {
-      GrammarEnumerationAlgorithm algorithm = (GrammarEnumerationAlgorithm)sender;
-      if (algorithm.Results.ContainsKey(BestTrainingModelResultName)) {
-        SymbolicRegressionModel model = (SymbolicRegressionModel)algorithm.Results[BestTrainingModelResultName].Value;
-        IRegressionSolution bestTrainingSolution = new RegressionSolution(model, algorithm.Problem.ProblemData);
+    private void OnStopped(object sender, EventArgs eventArgs) { }
 
-        algorithm.Results.AddOrUpdateResult(BestTrainingSolutionResultName, bestTrainingSolution);
-      }
+    private T GetValue<T>(IItem value) where T : struct {
+      var v = value as ValueTypeValue<T>;
+      if (v == null)
+        throw new ArgumentException(string.Format("Item is not of type {0}", typeof(ValueTypeValue<T>)));
+      return v.Value;
     }
 
     private void EvaluateSentence(GrammarEnumerationAlgorithm algorithm, SymbolString symbolString) {
+      var results = algorithm.Results;
+      var grammar = algorithm.Grammar;
       var problemData = algorithm.Problem.ProblemData;
 
@@ -74 +76 @@
 
       double r2 = Evaluate(problemData, tree, OptimizeConstants);
+      double bestR2 = results.ContainsKey(BestTrainingQualityResultName) ? GetValue<double>(results[BestTrainingQualityResultName].Value) : 0.0;
+      if (r2 < bestR2)
+        return;
 
-      double bestR2 = 0.0;
-      if (algorithm.Results.ContainsKey(BestTrainingQualityResultName))
-        bestR2 = ((DoubleValue)algorithm.Results[BestTrainingQualityResultName].Value).Value;
-      bool better = r2 > bestR2;
-      bool equallyGood = r2.IsAlmost(bestR2);
-      bool shorter = false;
+      var bestComplexity = int.MaxValue;
+      if (results.ContainsKey(BestComplexityResultName)) {
+        bestComplexity = GetValue<int>(results[BestComplexityResultName].Value);
+      } else if (algorithm.BestTrainingSentence != null) {
+        bestComplexity = grammar.GetComplexity(algorithm.BestTrainingSentence);
+        results.AddOrUpdateResult(BestComplexityResultName, new IntValue(bestComplexity));
+      }
+      var complexity = grammar.GetComplexity(symbolString);
 
-      if (!better && equallyGood) {
-        shorter = algorithm.BestTrainingSentence != null &&
-          algorithm.Grammar.GetComplexity(algorithm.BestTrainingSentence) > algorithm.Grammar.GetComplexity(symbolString);
-      }
-      if (better || (equallyGood && shorter)) {
-        algorithm.Results.AddOrUpdateResult(BestTrainingQualityResultName, new DoubleValue(r2));
-
-        SymbolicRegressionModel model = new SymbolicRegressionModel(
-          problemData.TargetVariable,
-          tree,
-          expressionTreeLinearInterpreter);
-
-        algorithm.Results.AddOrUpdateResult(BestTrainingModelResultName, model);
-
+      if (r2 > bestR2 || (r2.IsAlmost(bestR2) && complexity < bestComplexity)) {
+        results.AddOrUpdateResult(BestTrainingQualityResultName, new DoubleValue(r2));
+        results.AddOrUpdateResult(BestComplexityResultName, new IntValue(complexity));
         algorithm.BestTrainingSentence = symbolString;
       }
@@ -121 +117 @@
           }
         }
-
-        r2 = double.IsNaN(r2) ? 0.0 : r2;
-
       } else {
-        var target = problemData.TargetVariableTrainingValues;
-
-        SymbolicRegressionModel model = new SymbolicRegressionModel(
-          problemData.TargetVariable,
+        r2 = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(expressionTreeLinearInterpreter,
           tree,
-          expressionTreeLinearInterpreter);
-
-        var estVals = model.GetEstimatedValues(problemData.Dataset, problemData.TrainingIndices);
-        OnlineCalculatorError error;
-        r2 = OnlinePearsonsRCalculator.Calculate(target, estVals, out error);
-        if (error != OnlineCalculatorError.None) r2 = 0.0;
+          double.MinValue,
+          double.MaxValue,
+          problemData,
+          problemData.TrainingIndices,
+          applyLinearScaling: true);
       }
-
-      return r2;
+      return double.IsNaN(r2) ? 0.0 : r2;
     }
   }