source: branches/DataAnalysis/HeuristicLab.Problems.DataAnalysis.Regression/3.3/Symbolic/Analyzers/OverfittingAnalyzer.cs @ 4272

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

namespace HeuristicLab.Problems.DataAnalysis.Regression.Symbolic.Analyzers {
  [Item("OverfittingAnalyzer", "")]
  [StorableClass]
  public sealed class OverfittingAnalyzer : SingleSuccessorOperator, ISymbolicRegressionAnalyzer {
    private const string RandomParameterName = "Random";
    private const string SymbolicExpressionTreeParameterName = "SymbolicExpressionTree";
    private const string SymbolicExpressionTreeInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
    private const string ProblemDataParameterName = "ProblemData";
    private const string ValidationSamplesStartParameterName = "SamplesStart";
    private const string ValidationSamplesEndParameterName = "SamplesEnd";
    private const string UpperEstimationLimitParameterName = "UpperEstimationLimit";
    private const string LowerEstimationLimitParameterName = "LowerEstimationLimit";
    private const string EvaluatorParameterName = "Evaluator";
    private const string MaximizationParameterName = "Maximization";
    private const string RelativeNumberOfEvaluatedSamplesParameterName = "RelativeNumberOfEvaluatedSamples";

    #region parameter properties
    public ILookupParameter<IRandom> RandomParameter {
      get { return (ILookupParameter<IRandom>)Parameters[RandomParameterName]; }
    }
    public ScopeTreeLookupParameter<SymbolicExpressionTree> SymbolicExpressionTreeParameter {
      get { return (ScopeTreeLookupParameter<SymbolicExpressionTree>)Parameters[SymbolicExpressionTreeParameterName]; }
    }
    public ScopeTreeLookupParameter<DoubleValue> QualityParameter {
      get { return (ScopeTreeLookupParameter<DoubleValue>)Parameters["Quality"]; }
    }
    public IValueLookupParameter<ISymbolicExpressionTreeInterpreter> SymbolicExpressionTreeInterpreterParameter {
      get { return (IValueLookupParameter<ISymbolicExpressionTreeInterpreter>)Parameters[SymbolicExpressionTreeInterpreterParameterName]; }
    }
    public ILookupParameter<ISymbolicRegressionEvaluator> EvaluatorParameter {
      get { return (ILookupParameter<ISymbolicRegressionEvaluator>)Parameters[EvaluatorParameterName]; }
    }
    public ILookupParameter<BoolValue> MaximizationParameter {
      get { return (ILookupParameter<BoolValue>)Parameters[MaximizationParameterName]; }
    }
    public IValueLookupParameter<DataAnalysisProblemData> ProblemDataParameter {
      get { return (IValueLookupParameter<DataAnalysisProblemData>)Parameters[ProblemDataParameterName]; }
    }
    public IValueLookupParameter<IntValue> ValidationSamplesStartParameter {
      get { return (IValueLookupParameter<IntValue>)Parameters[ValidationSamplesStartParameterName]; }
    }
    public IValueLookupParameter<IntValue> ValidationSamplesEndParameter {
      get { return (IValueLookupParameter<IntValue>)Parameters[ValidationSamplesEndParameterName]; }
    }
    public IValueParameter<PercentValue> RelativeNumberOfEvaluatedSamplesParameter {
      get { return (IValueParameter<PercentValue>)Parameters[RelativeNumberOfEvaluatedSamplesParameterName]; }
    }

    public IValueLookupParameter<DoubleValue> UpperEstimationLimitParameter {
      get { return (IValueLookupParameter<DoubleValue>)Parameters[UpperEstimationLimitParameterName]; }
    }
    public IValueLookupParameter<DoubleValue> LowerEstimationLimitParameter {
      get { return (IValueLookupParameter<DoubleValue>)Parameters[LowerEstimationLimitParameterName]; }
    }
    public ILookupParameter<PercentValue> RelativeValidationQualityParameter {
      get { return (ILookupParameter<PercentValue>)Parameters["RelativeValidationQuality"]; }
    }
    //public IValueLookupParameter<PercentValue> RelativeValidationQualityLowerLimitParameter {
    //  get { return (IValueLookupParameter<PercentValue>)Parameters["RelativeValidationQualityLowerLimit"]; }
    //}
    //public IValueLookupParameter<PercentValue> RelativeValidationQualityUpperLimitParameter {
    //  get { return (IValueLookupParameter<PercentValue>)Parameters["RelativeValidationQualityUpperLimit"]; }
    //}
    public ILookupParameter<DoubleValue> TrainingValidationQualityCorrelationParameter {
      get { return (ILookupParameter<DoubleValue>)Parameters["TrainingValidationCorrelation"]; }
    }
    public IValueLookupParameter<DoubleValue> CorrelationLimitParameter {
      get { return (IValueLookupParameter<DoubleValue>)Parameters["CorrelationLimit"]; }
    }
    public ILookupParameter<BoolValue> OverfittingParameter {
      get { return (ILookupParameter<BoolValue>)Parameters["Overfitting"]; }
    }
    public ILookupParameter<ResultCollection> ResultsParameter {
      get { return (ILookupParameter<ResultCollection>)Parameters["Results"]; }
    }
    public ILookupParameter<DoubleValue> InitialTrainingQualityParameter {
      get { return (ILookupParameter<DoubleValue>)Parameters["InitialTrainingQuality"]; }
    }
    #endregion
    #region properties
    public IRandom Random {
      get { return RandomParameter.ActualValue; }
    }
    public ItemArray<SymbolicExpressionTree> SymbolicExpressionTree {
      get { return SymbolicExpressionTreeParameter.ActualValue; }
    }
    public ISymbolicExpressionTreeInterpreter SymbolicExpressionTreeInterpreter {
      get { return SymbolicExpressionTreeInterpreterParameter.ActualValue; }
    }
    public ISymbolicRegressionEvaluator Evaluator {
      get { return EvaluatorParameter.ActualValue; }
    }
    public BoolValue Maximization {
      get { return MaximizationParameter.ActualValue; }
    }
    public DataAnalysisProblemData ProblemData {
      get { return ProblemDataParameter.ActualValue; }
    }
    public IntValue ValidiationSamplesStart {
      get { return ValidationSamplesStartParameter.ActualValue; }
    }
    public IntValue ValidationSamplesEnd {
      get { return ValidationSamplesEndParameter.ActualValue; }
    }
    public PercentValue RelativeNumberOfEvaluatedSamples {
      get { return RelativeNumberOfEvaluatedSamplesParameter.Value; }
    }

    public DoubleValue UpperEstimationLimit {
      get { return UpperEstimationLimitParameter.ActualValue; }
    }
    public DoubleValue LowerEstimationLimit {
      get { return LowerEstimationLimitParameter.ActualValue; }
    }
    #endregion

    public OverfittingAnalyzer()
      : base() {
      Parameters.Add(new LookupParameter<IRandom>(RandomParameterName, "The random generator to use."));
      Parameters.Add(new LookupParameter<ISymbolicRegressionEvaluator>(EvaluatorParameterName, "The evaluator which should be used to evaluate the solution on the validation set."));
      Parameters.Add(new ScopeTreeLookupParameter<SymbolicExpressionTree>(SymbolicExpressionTreeParameterName, "The symbolic expression trees to analyze."));
      Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>("Quality"));
      Parameters.Add(new LookupParameter<BoolValue>(MaximizationParameterName, "The direction of optimization."));
      Parameters.Add(new ValueLookupParameter<ISymbolicExpressionTreeInterpreter>(SymbolicExpressionTreeInterpreterParameterName, "The interpreter that should be used for the analysis of symbolic expression trees."));
      Parameters.Add(new ValueLookupParameter<DataAnalysisProblemData>(ProblemDataParameterName, "The problem data for which the symbolic expression tree is a solution."));
      Parameters.Add(new ValueLookupParameter<IntValue>(ValidationSamplesStartParameterName, "The first index of the validation partition of the data set."));
      Parameters.Add(new ValueLookupParameter<IntValue>(ValidationSamplesEndParameterName, "The last index of the validation partition of the data set."));
      Parameters.Add(new ValueParameter<PercentValue>(RelativeNumberOfEvaluatedSamplesParameterName, "The relative number of samples of the dataset partition, which should be randomly chosen for evaluation between the start and end index.", new PercentValue(1)));
      Parameters.Add(new ValueLookupParameter<DoubleValue>(UpperEstimationLimitParameterName, "The upper estimation limit that was set for the evaluation of the symbolic expression trees."));
      Parameters.Add(new ValueLookupParameter<DoubleValue>(LowerEstimationLimitParameterName, "The lower estimation limit that was set for the evaluation of the symbolic expression trees."));
      Parameters.Add(new LookupParameter<PercentValue>("RelativeValidationQuality"));
      //Parameters.Add(new ValueLookupParameter<PercentValue>("RelativeValidationQualityUpperLimit", new PercentValue(0.05)));
      //Parameters.Add(new ValueLookupParameter<PercentValue>("RelativeValidationQualityLowerLimit", new PercentValue(-0.05)));
      Parameters.Add(new LookupParameter<DoubleValue>("TrainingValidationCorrelation"));
      Parameters.Add(new ValueLookupParameter<DoubleValue>("CorrelationLimit", new DoubleValue(0.65)));
      Parameters.Add(new LookupParameter<BoolValue>("Overfitting"));
      Parameters.Add(new LookupParameter<ResultCollection>("Results"));
      Parameters.Add(new LookupParameter<DoubleValue>("InitialTrainingQuality"));
    }

    [StorableConstructor]
    private OverfittingAnalyzer(bool deserializing) : base(deserializing) { }

    [StorableHook(HookType.AfterDeserialization)]
    private void AfterDeserialization() {
      if (!Parameters.ContainsKey("InitialTrainingQuality")) {
        Parameters.Add(new LookupParameter<DoubleValue>("InitialTrainingQuality"));
      }
      //if (!Parameters.ContainsKey("RelativeValidationQualityUpperLimit")) {
      //  Parameters.Add(new ValueLookupParameter<PercentValue>("RelativeValidationQualityUpperLimit", new PercentValue(0.05)));
      //}
      //if (!Parameters.ContainsKey("RelativeValidationQualityLowerLimit")) {
      //  Parameters.Add(new ValueLookupParameter<PercentValue>("RelativeValidationQualityLowerLimit", new PercentValue(-0.05)));
      //}
    }

    public override IOperation Apply() {
      var trees = SymbolicExpressionTree;
      ItemArray<DoubleValue> qualities = QualityParameter.ActualValue;

      string targetVariable = ProblemData.TargetVariable.Value;

      // select a random subset of rows in the validation set
      int validationStart = ValidiationSamplesStart.Value;
      int validationEnd = ValidationSamplesEnd.Value;
      int seed = Random.Next();
      int count = (int)((validationEnd - validationStart) * RelativeNumberOfEvaluatedSamples.Value);
      if (count == 0) count = 1;
      IEnumerable<int> rows = RandomEnumerable.SampleRandomNumbers(seed, validationStart, validationEnd, count);

      double upperEstimationLimit = UpperEstimationLimit != null ? UpperEstimationLimit.Value : double.PositiveInfinity;
      double lowerEstimationLimit = LowerEstimationLimit != null ? LowerEstimationLimit.Value : double.NegativeInfinity;

      //double bestQuality = Maximization.Value ? double.NegativeInfinity : double.PositiveInfinity;
      //SymbolicExpressionTree bestTree = null;

      List<double> validationQualities = new List<double>();
      foreach (var tree in trees) {
        double quality = Evaluator.Evaluate(SymbolicExpressionTreeInterpreter, tree,
          lowerEstimationLimit, upperEstimationLimit,
          ProblemData.Dataset, targetVariable,
         rows);
        validationQualities.Add(quality);
        //if ((Maximization.Value && quality > bestQuality) ||
        //    (!Maximization.Value && quality < bestQuality)) {
        //  bestQuality = quality;
        //  bestTree = tree;
        //}
      }

      //if (RelativeValidationQualityParameter.ActualValue == null) {
      // first call initialize the relative quality using the difference between average training and validation quality
      double avgTrainingQuality = qualities.Select(x => x.Value).Median();
      double avgValidationQuality = validationQualities.Median();

      if (Maximization.Value)
        RelativeValidationQualityParameter.ActualValue = new PercentValue(avgValidationQuality / avgTrainingQuality - 1);
      else {
        RelativeValidationQualityParameter.ActualValue = new PercentValue(avgTrainingQuality / avgValidationQuality - 1);
      }
      //}

      // cut away 0.0 values to make the correlation stronger
      // necessary because R² values of 0.0 are strong outliers
      //int percentile = (int)Math.Round(0.1 * validationQualities.Count);
      //double validationCutOffValue = validationQualities.OrderBy(x => x).ElementAt(percentile);
      //double trainingCutOffValue = qualities.Select(x => x.Value).OrderBy(x => x).ElementAt(percentile);
      double validationCutOffValue = 0.05;
      double trainingCutOffValue = validationCutOffValue;

      double[] validationArr = new double[validationQualities.Count];
      double[] trainingArr = new double[validationQualities.Count];
      int arrIndex = 0;
      for (int i = 0; i < validationQualities.Count; i++) {
        if (validationQualities[i] > validationCutOffValue &&
            qualities[i].Value > trainingCutOffValue) {
          validationArr[arrIndex] = validationQualities[i];
          trainingArr[arrIndex] = qualities[i].Value;
          arrIndex++;
        }
      }
      double r = alglib.correlation.spearmanrankcorrelation(trainingArr, validationArr, arrIndex);
      TrainingValidationQualityCorrelationParameter.ActualValue = new DoubleValue(r);
      if (InitialTrainingQualityParameter.ActualValue == null)
        InitialTrainingQualityParameter.ActualValue = new DoubleValue(avgValidationQuality);
      bool overfitting =
        avgTrainingQuality > InitialTrainingQualityParameter.ActualValue.Value &&  // better on training than in initial generation
        r < CorrelationLimitParameter.ActualValue.Value;  // low correlation between training and validation quality

      //// if validation quality is within a certain margin of percentage deviation (default -5% .. 5%) then there is no overfitting
      //// correlation is also bad when underfitting but validation quality cannot be a lot larger than training quality if overfitting
      //(RelativeValidationQualityParameter.ActualValue.Value > RelativeValidationQualityUpperLimitParameter.ActualValue.Value || // better on training than on validation
      // RelativeValidationQualityParameter.ActualValue.Value < RelativeValidationQualityLowerLimitParameter.ActualValue.Value); // better on training than on validation

      OverfittingParameter.ActualValue = new BoolValue(overfitting);
      return base.Apply();
    }

    [StorableHook(HookType.AfterDeserialization)]
    private void Initialize() { }

    private static void AddValue(DataTable table, double data, string name, string description) {
      DataRow row;
      table.Rows.TryGetValue(name, out row);
      if (row == null) {
        row = new DataRow(name, description);
        row.Values.Add(data);
        table.Rows.Add(row);
      } else {
        row.Values.Add(data);
      }
    }
  }
}