source: trunk/sources/HeuristicLab.Problems.DataAnalysis.Regression/3.3/Symbolic/BestValidationSymbolicRegressionSolutionVisualizer.cs @ 3559

#region License Information
/* HeuristicLab
 * Copyright (C) 2002-2010 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
 *
 * This file is part of HeuristicLab.
 *
 * HeuristicLab is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * HeuristicLab is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
 */
#endregion

using System.Linq;
using HeuristicLab.Common;
using HeuristicLab.Core;
using HeuristicLab.Data;
using HeuristicLab.Operators;
using HeuristicLab.Optimization;
using HeuristicLab.Parameters;
using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
using HeuristicLab.Problems.DataAnalysis.Evaluators;
using HeuristicLab.Problems.DataAnalysis.Symbolic;
using System.Collections.Generic;
using HeuristicLab.Analysis;

using HeuristicLab.Problems.DataAnalysis.Symbolic.Symbols;

namespace HeuristicLab.Problems.DataAnalysis.Regression.Symbolic {
  /// <summary>
  /// An operator for visualizing the best symbolic regression solution based on the validation set.
  /// </summary>
  [Item("BestSymbolicExpressionTreeVisualizer", "An operator for visualizing the best symbolic regression solution based on the validation set.")]
  [StorableClass]
  public sealed class BestValidationSymbolicRegressionSolutionVisualizer : SingleSuccessorOperator, ISingleObjectiveSolutionsVisualizer, ISolutionsVisualizer {
    private const string SymbolicExpressionTreeInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
    private const string UpperEstimationLimitParameterName = "UpperEstimationLimit";
    private const string LowerEstimationLimitParameterName = "LowerEstimationLimit";
    private const string AlphaParameterName = "Alpha";
    private const string BetaParameterName = "Beta";
    private const string SymbolicRegressionModelParameterName = "SymbolicRegressionModel";
    private const string DataAnalysisProblemDataParameterName = "DataAnalysisProblemData";
    private const string BestValidationSolutionParameterName = "BestValidationSolution";
    private const string ValidationSamplesStartParameterName = "ValidationSamplesStart";
    private const string ValidationSamplesEndParameterName = "ValidationSamplesEnd";
    private const string QualityParameterName = "Quality";
    private const string ResultsParameterName = "Results";
    private const string VariableFrequenciesParameterName = "VariableFrequencies";

    #region parameter properties
    public ILookupParameter<ISymbolicExpressionTreeInterpreter> SymbolicExpressionTreeInterpreterParameter {
      get { return (ILookupParameter<ISymbolicExpressionTreeInterpreter>)Parameters[SymbolicExpressionTreeInterpreterParameterName]; }
    }
    public IValueLookupParameter<DoubleValue> UpperEstimationLimitParameter {
      get { return (IValueLookupParameter<DoubleValue>)Parameters[UpperEstimationLimitParameterName]; }
    }
    public IValueLookupParameter<DoubleValue> LowerEstimationLimitParameter {
      get { return (IValueLookupParameter<DoubleValue>)Parameters[LowerEstimationLimitParameterName]; }
    }
    public IValueLookupParameter<IntValue> ValidationSamplesStartParameter {
      get { return (IValueLookupParameter<IntValue>)Parameters[ValidationSamplesStartParameterName]; }
    }
    public IValueLookupParameter<IntValue> ValidationSamplesEndParameter {
      get { return (IValueLookupParameter<IntValue>)Parameters[ValidationSamplesEndParameterName]; }
    }

    public ILookupParameter<ItemArray<SymbolicExpressionTree>> SymbolicExpressionTreeParameter {
      get { return (ILookupParameter<ItemArray<SymbolicExpressionTree>>)Parameters[SymbolicRegressionModelParameterName]; }
    }
    public ILookupParameter<ItemArray<DoubleValue>> AlphaParameter {
      get { return (ILookupParameter<ItemArray<DoubleValue>>)Parameters[AlphaParameterName]; }
    }
    public ILookupParameter<ItemArray<DoubleValue>> BetaParameter {
      get { return (ILookupParameter<ItemArray<DoubleValue>>)Parameters[BetaParameterName]; }
    }
    public ILookupParameter<DataAnalysisProblemData> DataAnalysisProblemDataParameter {
      get { return (ILookupParameter<DataAnalysisProblemData>)Parameters[DataAnalysisProblemDataParameterName]; }
    }
    public ILookupParameter<SymbolicRegressionSolution> BestValidationSolutionParameter {
      get { return (ILookupParameter<SymbolicRegressionSolution>)Parameters[BestValidationSolutionParameterName]; }
    }
    ILookupParameter ISolutionsVisualizer.VisualizationParameter {
      get { return BestValidationSolutionParameter; }
    }

    public ILookupParameter<ItemArray<DoubleValue>> QualityParameter {
      get { return (ILookupParameter<ItemArray<DoubleValue>>)Parameters[QualityParameterName]; }
    }

    public ILookupParameter<ResultCollection> ResultParameter {
      get { return (ILookupParameter<ResultCollection>)Parameters[ResultsParameterName]; }
    }
    public ILookupParameter<DataTable> VariableFrequenciesParameter {
      get { return (ILookupParameter<DataTable>)Parameters[VariableFrequenciesParameterName]; }
    }

    #endregion

    #region properties
    public ISymbolicExpressionTreeInterpreter SymbolicExpressionTreeInterpreter {
      get { return SymbolicExpressionTreeInterpreterParameter.ActualValue; }
    }
    public DoubleValue UpperEstimationLimit {
      get { return UpperEstimationLimitParameter.ActualValue; }
    }
    public DoubleValue LowerEstimationLimit {
      get { return LowerEstimationLimitParameter.ActualValue; }
    }
    public IntValue ValidationSamplesStart {
      get { return ValidationSamplesStartParameter.ActualValue; }
    }
    public IntValue ValidationSamplesEnd {
      get { return ValidationSamplesEndParameter.ActualValue; }
    }
    public DataTable VariableFrequencies {
      get { return VariableFrequenciesParameter.ActualValue; }
      set { VariableFrequenciesParameter.ActualValue = value; }
    }
    #endregion

    public BestValidationSymbolicRegressionSolutionVisualizer()
      : base() {
      Parameters.Add(new SubScopesLookupParameter<SymbolicExpressionTree>(SymbolicRegressionModelParameterName, "The symbolic regression solutions from which the best solution should be visualized."));
      Parameters.Add(new SubScopesLookupParameter<DoubleValue>(QualityParameterName, "The quality of the symbolic regression solutions."));
      Parameters.Add(new LookupParameter<DataAnalysisProblemData>(DataAnalysisProblemDataParameterName, "The symbolic regression problme data on which the best solution should be evaluated."));
      Parameters.Add(new LookupParameter<ISymbolicExpressionTreeInterpreter>(SymbolicExpressionTreeInterpreterParameterName, "The interpreter that should be used to calculate the output values of symbolic expression trees."));
      Parameters.Add(new SubScopesLookupParameter<DoubleValue>(AlphaParameterName, "Alpha parameter for linear scaling of the estimated values."));
      Parameters.Add(new SubScopesLookupParameter<DoubleValue>(BetaParameterName, "Beta parameter for linear scaling ot the estimated values."));
      Parameters.Add(new ValueLookupParameter<DoubleValue>(UpperEstimationLimitParameterName, "The upper limit that should be used as cut off value for the output values of symbolic expression trees."));
      Parameters.Add(new ValueLookupParameter<DoubleValue>(LowerEstimationLimitParameterName, "The lower limit that should be used as cut off value for the output values of symbolic expression trees."));
      Parameters.Add(new ValueLookupParameter<IntValue>(ValidationSamplesStartParameterName, "The start index of the validation partition (part of the training partition)."));
      Parameters.Add(new ValueLookupParameter<IntValue>(ValidationSamplesEndParameterName, "The end index of the validation partition (part of the training partition)."));
      Parameters.Add(new LookupParameter<SymbolicRegressionSolution>(BestValidationSolutionParameterName, "The best symbolic expression tree based on the validation data for the symbolic regression problem."));
      Parameters.Add(new LookupParameter<DataTable>(VariableFrequenciesParameterName, "The relative variable reference frequencies aggregated over the whole population."));
      Parameters.Add(new LookupParameter<ResultCollection>(ResultsParameterName, "The result collection of the algorithm."));
    }

    public override IOperation Apply() {
      ItemArray<SymbolicExpressionTree> expressions = SymbolicExpressionTreeParameter.ActualValue;
      ItemArray<DoubleValue> alphas = AlphaParameter.ActualValue;
      ItemArray<DoubleValue> betas = BetaParameter.ActualValue;
      var scaledExpressions = from i in Enumerable.Range(0, expressions.Count())
                              let expr = expressions[i]
                              let alpha = alphas[i] == null ? 0.0 : alphas[i].Value
                              let beta = betas[i] == null ? 1.0 : betas[i].Value
                              select new { Expression = expr, Alpha = alpha, Beta = beta };
      DataAnalysisProblemData problemData = DataAnalysisProblemDataParameter.ActualValue;
      #region update variable frequencies
      var inputVariables = problemData.InputVariables.Select(x => x.Value);
      if (VariableFrequencies == null) {
        VariableFrequencies = new DataTable("Variable Frequencies", "Relative frequency of variable references aggregated over the whole population.");
        AddResult("VariableFrequencies", VariableFrequencies);
        // add a data row for each input variable
        foreach (var inputVariable in inputVariables)
          VariableFrequencies.Rows.Add(new DataRow(inputVariable));
      }
      foreach (var pair in VariableFrequencyAnalyser.CalculateVariableFrequencies(expressions, inputVariables)) {
        VariableFrequencies.Rows[pair.Key].Values.Add(pair.Value);
      }
      #endregion

      #region determination of validation-best solution
      int validationSamplesStart = ValidationSamplesStart.Value;
      int validationSamplesEnd = ValidationSamplesEnd.Value;
      var validationValues = problemData.Dataset.GetVariableValues(problemData.TargetVariable.Value, validationSamplesStart, validationSamplesEnd);
      double upperEstimationLimit = UpperEstimationLimit.Value;
      double lowerEstimationLimit = LowerEstimationLimit.Value;
      var currentBestExpression = (from expression in scaledExpressions
                                   let validationQuality =
                                     SymbolicRegressionScaledMeanSquaredErrorEvaluator.CalculateWithScaling(
                                       SymbolicExpressionTreeInterpreter, expression.Expression,
                                       lowerEstimationLimit, upperEstimationLimit,
                                       problemData.Dataset, problemData.TargetVariable.Value,
                                       validationSamplesStart, validationSamplesEnd,
                                       expression.Beta, expression.Alpha)
                                   select new { Expression = expression, ValidationQuality = validationQuality })
                                   .OrderBy(x => x.ValidationQuality)
                                   .First();

      SymbolicRegressionSolution bestOfRunSolution = BestValidationSolutionParameter.ActualValue;
      #endregion
      #region update of validation-best solution
      if (bestOfRunSolution == null) {
        // no best of run solution yet -> make a solution from the currentBestExpression
        UpdateBestOfRunSolution(problemData, currentBestExpression.Expression.Expression, SymbolicExpressionTreeInterpreter, lowerEstimationLimit, upperEstimationLimit, currentBestExpression.Expression.Alpha, currentBestExpression.Expression.Beta);
      } else {
        // compare quality of current best with best of run solution
        var estimatedValidationValues = bestOfRunSolution.EstimatedValues.Skip(validationSamplesStart).Take(validationSamplesEnd - validationSamplesStart);
        var bestOfRunValidationQuality = SimpleMSEEvaluator.Calculate(validationValues, estimatedValidationValues);
        if (bestOfRunValidationQuality > currentBestExpression.ValidationQuality) {
          UpdateBestOfRunSolution(problemData, currentBestExpression.Expression.Expression, SymbolicExpressionTreeInterpreter, lowerEstimationLimit, upperEstimationLimit, currentBestExpression.Expression.Alpha, currentBestExpression.Expression.Beta);
        }
      }
      #endregion
      return base.Apply();
    }

    private void UpdateBestOfRunSolution(DataAnalysisProblemData problemData, SymbolicExpressionTree tree, ISymbolicExpressionTreeInterpreter interpreter,
      double lowerEstimationLimit, double upperEstimationLimit,
      double alpha, double beta) {
      var newBestSolution = CreateDataAnalysisSolution(problemData, tree, interpreter, lowerEstimationLimit, upperEstimationLimit, alpha, beta);
      if (BestValidationSolutionParameter.ActualValue == null)
        BestValidationSolutionParameter.ActualValue = newBestSolution;
      else
        // only update model
        BestValidationSolutionParameter.ActualValue.Model = newBestSolution.Model;

      AddResult("NumberOfInputVariables", new IntValue(CountInputVariables(tree)));

      var trainingValues = problemData.Dataset.GetVariableValues(problemData.TargetVariable.Value, problemData.TrainingSamplesStart.Value, problemData.TrainingSamplesEnd.Value);
      var testValues = problemData.Dataset.GetVariableValues(problemData.TargetVariable.Value, problemData.TestSamplesStart.Value, problemData.TestSamplesEnd.Value);

      AddResult("MeanSquaredError (Training)", new DoubleValue(SimpleMSEEvaluator.Calculate(trainingValues, newBestSolution.EstimatedTrainingValues)));
      AddResult("MeanRelativeError (Training)", new PercentValue(SimpleMeanAbsolutePercentageErrorEvaluator.Calculate(trainingValues, newBestSolution.EstimatedTrainingValues)));
      AddResult("RSquared (Training)", new DoubleValue(SimpleRSquaredEvaluator.Calculate(trainingValues, newBestSolution.EstimatedTrainingValues)));

      AddResult("MeanSquaredError (Test)", new DoubleValue(SimpleMSEEvaluator.Calculate(testValues, newBestSolution.EstimatedTestValues)));
      AddResult("MeanRelativeError (Test)", new PercentValue(SimpleMeanAbsolutePercentageErrorEvaluator.Calculate(testValues, newBestSolution.EstimatedTestValues)));
      AddResult("RSquared (Test)", new DoubleValue(SimpleRSquaredEvaluator.Calculate(testValues, newBestSolution.EstimatedTestValues)));
    }

    private int CountInputVariables(SymbolicExpressionTree tree) {
      return (from node in tree.IterateNodesPrefix().OfType<VariableTreeNode>()
              select node.VariableName)
             .Distinct()
             .Count();
    }

    private void AddResult(string resultName, IItem value) {
      var resultCollection = ResultParameter.ActualValue;
      if (resultCollection.ContainsKey(resultName)) {
        resultCollection[resultName].Value = value;
      } else {
        resultCollection.Add(new Result(resultName, value));
      }
    }

    private SymbolicRegressionSolution CreateDataAnalysisSolution(DataAnalysisProblemData problemData, SymbolicExpressionTree tree, ISymbolicExpressionTreeInterpreter interpreter,
      double lowerEstimationLimit, double upperEstimationLimit,
      double alpha, double beta) {
      var mainBranch = tree.Root.SubTrees[0].SubTrees[0];
      var scaledMainBranch = MakeSum(MakeProduct(beta, mainBranch), alpha);

      // remove the main branch before cloning to prevent cloning of sub-trees
      tree.Root.SubTrees[0].RemoveSubTree(0);
      var scaledTree = (SymbolicExpressionTree)tree.Clone();
      // insert main branch into the original tree again 
      tree.Root.SubTrees[0].InsertSubTree(0, mainBranch);
      // insert the scaled main branch into the cloned tree
      scaledTree.Root.SubTrees[0].InsertSubTree(0, scaledMainBranch);
      // create a new solution using the scaled tree
      var model = new SymbolicRegressionModel(interpreter, scaledTree, problemData.InputVariables.Select(s => s.Value));
      return new SymbolicRegressionSolution(problemData, model, lowerEstimationLimit, upperEstimationLimit);
    }

    private SymbolicExpressionTreeNode MakeSum(SymbolicExpressionTreeNode treeNode, double alpha) {
      var node = (new Addition()).CreateTreeNode();
      var alphaConst = MakeConstant(alpha);
      node.AddSubTree(treeNode);
      node.AddSubTree(alphaConst);
      return node;
    }

    private SymbolicExpressionTreeNode MakeProduct(double beta, SymbolicExpressionTreeNode treeNode) {
      var node = (new Multiplication()).CreateTreeNode();
      var betaConst = MakeConstant(beta);
      node.AddSubTree(treeNode);
      node.AddSubTree(betaConst);
      return node;
    }

    private SymbolicExpressionTreeNode MakeConstant(double c) {
      var node = (ConstantTreeNode)(new Constant()).CreateTreeNode();
      node.Value = c;
      return node;
    }
  }
}