source: branches/2886_SymRegGrammarEnumeration/HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration/Analysis/RSquaredEvaluator.cs @ 15910

using System;
using System.Diagnostics;
using HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration.GrammarEnumeration;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
using HeuristicLab.Optimization;
using HeuristicLab.Problems.DataAnalysis;
using HeuristicLab.Problems.DataAnalysis.Symbolic;
using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;

namespace HeuristicLab.Algorithms.DataAnalysis.SymRegGrammarEnumeration {
  public class RSquaredEvaluator : Item, IGrammarEnumerationAnalyzer {
    public static readonly string BestTrainingQualityResultName = "Best R² (Training)";
    public static readonly string BestTrainingModelResultName = "Best model (Training)";
    public static readonly string BestTrainingSolutionResultName = "Best solution (Training)";

    private static readonly ISymbolicDataAnalysisExpressionTreeInterpreter expressionTreeLinearInterpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();

    public bool OptimizeConstants { get; set; }

    public RSquaredEvaluator() { }

    protected RSquaredEvaluator(RSquaredEvaluator original, Cloner cloner) : base(original, cloner) {
      this.OptimizeConstants = original.OptimizeConstants;
    }

    public override IDeepCloneable Clone(Cloner cloner) {
      return new RSquaredEvaluator(this, cloner);
    }

    public void Register(GrammarEnumerationAlgorithm algorithm) {
      algorithm.Started += OnStarted;
      algorithm.Stopped += OnStopped;

      algorithm.DistinctSentenceGenerated += AlgorithmOnDistinctSentenceGenerated;
    }

    public void Deregister(GrammarEnumerationAlgorithm algorithm) {
      algorithm.Started -= OnStarted;
      algorithm.Stopped -= OnStopped;

      algorithm.DistinctSentenceGenerated -= AlgorithmOnDistinctSentenceGenerated;
    }

    private void AlgorithmOnDistinctSentenceGenerated(object sender, PhraseAddedEventArgs phraseAddedEventArgs) {
      GrammarEnumerationAlgorithm algorithm = (GrammarEnumerationAlgorithm)sender;
      EvaluateSentence(algorithm, phraseAddedEventArgs.NewPhrase);
    }

    private void OnStarted(object sender, EventArgs eventArgs) {
      GrammarEnumerationAlgorithm algorithm = (GrammarEnumerationAlgorithm)sender;

      algorithm.BestTrainingSentence = null;
    }

    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);

        algorithm.Results.AddOrUpdateResult(BestTrainingSolutionResultName, bestTrainingSolution);
      }
    }

    private void EvaluateSentence(GrammarEnumerationAlgorithm algorithm, SymbolString symbolString) {
      var problemData = algorithm.Problem.ProblemData;

      SymbolicExpressionTree tree = algorithm.Grammar.ParseSymbolicExpressionTree(symbolString);
      Debug.Assert(SymbolicRegressionConstantOptimizationEvaluator.CanOptimizeConstants(tree));

      double r2 = Evaluate(problemData, tree, OptimizeConstants);

      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;

      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);

        algorithm.BestTrainingSentence = symbolString;
      }
    }

    public static double Evaluate(IRegressionProblemData problemData, SymbolicExpressionTree tree, bool optimizeConstants = true) {
      double r2;

      // TODO: Initialize constant values randomly
      // TODO: Restarts
      if (optimizeConstants) {
        r2 = SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(expressionTreeLinearInterpreter,
          tree,
          problemData,
          problemData.TrainingIndices,
          applyLinearScaling: false,
          maxIterations: 50,
          updateVariableWeights: true,
          updateConstantsInTree: true);

        foreach (var symbolicExpressionTreeNode in tree.IterateNodesPostfix()) {
          ConstantTreeNode constTreeNode = symbolicExpressionTreeNode as ConstantTreeNode;
          if (constTreeNode != null && constTreeNode.Value.IsAlmost(0.0)) {
            constTreeNode.Value = 0.0;
          }
        }

      } else {
        var target = problemData.TargetVariableTrainingValues;

        SymbolicRegressionModel model = new SymbolicRegressionModel(
          problemData.TargetVariable,
          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;
      }

      return r2;
    }
  }
}