Free cookie consent management tool by TermsFeed Policy Generator

Changeset 13665


Ignore:
Timestamp:
03/08/16 11:33:21 (9 years ago)
Author:
bburlacu
Message:

#2288: Improve calculation of variable impacts in the analyzer

File:
1 edited

Legend:

Unmodified
Added
Removed

  • TabularUnified branches/HeuristicLab.VariableInteractionNetworks/HeuristicLab.VariableInteractionNetworks/3.3/SymbolicDataAnalysisVariableImpactsAnalyzer.cs

    r12568 r13665  
    1 using System;
     1#region License Information
     2/* HeuristicLab
     3 * Copyright (C) 2002-2016 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
     4 *
     5 * This file is part of HeuristicLab.
     6 *
     7 * HeuristicLab is free software: you can redistribute it and/or modify
     8 * it under the terms of the GNU General Public License as published by
     9 * the Free Software Foundation, either version 3 of the License, or
     10 * (at your option) any later version.
     11 *
     12 * HeuristicLab is distributed in the hope that it will be useful,
     13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
     14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
     15 * GNU General Public License for more details.
     16 *
     17 * You should have received a copy of the GNU General Public License
     18 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
     19 */
     20#endregion
     21
     22using System;
    223using System.Linq;
    324using HeuristicLab.Analysis;
    … …  
    526using HeuristicLab.Core;
    627using HeuristicLab.Data;
     28using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
    729using HeuristicLab.Optimization;
    830using HeuristicLab.Parameters;
    … …  
    1133using HeuristicLab.Problems.DataAnalysis.Symbolic;
    1234using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
    13 using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
    14 
    15 namespace HeuristicLab.VariableInteractionNetworks
    16 {
    17     [Item("SymbolicRegressionVariableImpactsAnalyzer", "An analyzer which calculates variable impacts based on the average node impacts from the tree")]
    18     [StorableClass]
    19     public class SymbolicRegressionVariableImpactsAnalyzer : SymbolicDataAnalysisAnalyzer
    20     {
    21         private const string UpdateCounterParameterName = "UpdateCounter";
    22         private const string UpdateIntervalParameterName = "UpdateInterval";
    23         public const string QualityParameterName = "Quality";
    24         private const string SymbolicDataAnalysisTreeInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
    25         private const string ProblemDataParameterName = "ProblemData";
    26         private const string ApplyLinearScalingParameterName = "ApplyLinearScaling";
    27         private const string MaxCOIterationsParameterName = "MaxCOIterations";
    28         private const string EstimationLimitsParameterName = "EstimationLimits";
    29         private const string EvaluatorParameterName = "Evaluator";
    30 
    31         private const string VariableFrequenciesParameterName = "VariableFrequencies";
    32         private const string VariableImpactsParameterName = "AverageVariableImpacts";
    33         private const string PercentageBestParameterName = "PercentageBest";
    34         private const string LastGenerationsParameterName = "LastGenerations";
    35         private const string MaximumGenerationsParameterName = "MaximumGenerations";
    36         private const string OptimizeConstantsParameterName = "OptimizeConstants";
    37         private const string PruneTreesParameterName = "PruneTrees";
    38 
    39         private SymbolicDataAnalysisExpressionTreeSimplifier simplifier;
    40         private SymbolicRegressionSolutionImpactValuesCalculator impactsCalculator;
    41 
    42         #region parameters
    43         public ValueParameter<IntValue> UpdateCounterParameter
    44         {
    45             get { return (ValueParameter<IntValue>)Parameters[UpdateCounterParameterName]; }
    46         }
    47         public ValueParameter<IntValue> UpdateIntervalParameter
    48         {
    49             get { return (ValueParameter<IntValue>)Parameters[UpdateIntervalParameterName]; }
    50         }
    51         public IScopeTreeLookupParameter<DoubleValue> QualityParameter
    52         {
    53             get { return (IScopeTreeLookupParameter<DoubleValue>)Parameters[QualityParameterName]; }
    54         }
    55         public ILookupParameter<ISymbolicDataAnalysisExpressionTreeInterpreter> SymbolicDataAnalysisTreeInterpreterParameter
    56         {
    57             get { return (ILookupParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>)Parameters[SymbolicDataAnalysisTreeInterpreterParameterName]; }
    58         }
    59         public ILookupParameter<IRegressionProblemData> ProblemDataParameter
    60         {
    61             get { return (ILookupParameter<IRegressionProblemData>)Parameters[ProblemDataParameterName]; }
    62         }
    63         public ILookupParameter<BoolValue> ApplyLinearScalingParameter
    64         {
    65             get { return (ILookupParameter<BoolValue>)Parameters[ApplyLinearScalingParameterName]; }
    66         }
    67         public IFixedValueParameter<IntValue> MaxCOIterationsParameter
    68         {
    69             get { return (IFixedValueParameter<IntValue>)Parameters[MaxCOIterationsParameterName]; }
    70         }
    71         public ILookupParameter<DoubleLimit> EstimationLimitsParameter
    72         {
    73             get { return (ILookupParameter<DoubleLimit>)Parameters[EstimationLimitsParameterName]; }
    74         }
    75         public ILookupParameter<ISymbolicRegressionSingleObjectiveEvaluator> EvaluatorParameter
    76         {
    77             get { return (ILookupParameter<ISymbolicRegressionSingleObjectiveEvaluator>)Parameters[EvaluatorParameterName]; }
    78         }
    79         public ILookupParameter<DataTable> VariableImpactsParameter
    80         {
    81             get { return (ILookupParameter<DataTable>)Parameters[VariableImpactsParameterName]; }
    82         }
    83         public IFixedValueParameter<PercentValue> PercentageBestParameter
    84         {
    85             get { return (IFixedValueParameter<PercentValue>)Parameters[PercentageBestParameterName]; }
    86         }
    87         public IFixedValueParameter<IntValue> LastGenerationsParameter
    88         {
    89             get { return (IFixedValueParameter<IntValue>)Parameters[LastGenerationsParameterName]; }
    90         }
    91         public IFixedValueParameter<BoolValue> OptimizeConstantsParameter
    92         {
    93             get { return (IFixedValueParameter<BoolValue>)Parameters[OptimizeConstantsParameterName]; }
    94         }
    95         public IFixedValueParameter<BoolValue> PruneTreesParameter
    96         {
    97             get { return (IFixedValueParameter<BoolValue>)Parameters[PruneTreesParameterName]; }
    98         }
    99         private ILookupParameter<IntValue> MaximumGenerationsParameter
    100         {
    101             get { return (ILookupParameter<IntValue>)Parameters[MaximumGenerationsParameterName]; }
    102         }
    103         #endregion
    104 
    105         #region parameter properties
    106         public int UpdateCounter
    107         {
    108             get { return UpdateCounterParameter.Value.Value; }
    109             set { UpdateCounterParameter.Value.Value = value; }
    110         }
    111         public int UpdateInterval
    112         {
    113             get { return UpdateIntervalParameter.Value.Value; }
    114             set { UpdateIntervalParameter.Value.Value = value; }
    115         }
    116         #endregion
    117 
    118         public SymbolicRegressionVariableImpactsAnalyzer()
    119         {
    120             Parameters.Add(new ValueParameter<IntValue>(UpdateCounterParameterName, new IntValue(0)));
    121             Parameters.Add(new ValueParameter<IntValue>(UpdateIntervalParameterName, new IntValue(1)));
    122             Parameters.Add(new LookupParameter<IRegressionProblemData>(ProblemDataParameterName));
    123             Parameters.Add(new LookupParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>(SymbolicDataAnalysisTreeInterpreterParameterName));
    124             Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>(QualityParameterName, "The individual qualities."));
    125             Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName));
    126             Parameters.Add(new LookupParameter<DoubleLimit>(EstimationLimitsParameterName));
    127             Parameters.Add(new FixedValueParameter<IntValue>(MaxCOIterationsParameterName));
    128             Parameters.Add(new LookupParameter<DataTable>(VariableImpactsParameterName, "The relative variable relevance calculated as the average relative variable frequency over the whole run."));
    129             Parameters.Add(new FixedValueParameter<PercentValue>(PercentageBestParameterName));
    130             Parameters.Add(new FixedValueParameter<IntValue>(LastGenerationsParameterName));
    131             Parameters.Add(new FixedValueParameter<BoolValue>(OptimizeConstantsParameterName, new BoolValue(false)));
    132             Parameters.Add(new FixedValueParameter<BoolValue>(PruneTreesParameterName, new BoolValue(false)));
    133             Parameters.Add(new LookupParameter<IntValue>(MaximumGenerationsParameterName, "The maximum number of generations which should be processed."));
    134 
    135             impactsCalculator = new SymbolicRegressionSolutionImpactValuesCalculator();
    136             simplifier = new SymbolicDataAnalysisExpressionTreeSimplifier();
    137         }
    138 
    139         [StorableConstructor]
    140         protected SymbolicRegressionVariableImpactsAnalyzer(bool deserializing) : base(deserializing) { }
    141 
    142         [StorableHook(HookType.AfterDeserialization)]
    143         private void AfterDeserialization()
    144         {
    145             impactsCalculator = new SymbolicRegressionSolutionImpactValuesCalculator();
    146             simplifier = new SymbolicDataAnalysisExpressionTreeSimplifier();
    147         }
    148 
    149         protected SymbolicRegressionVariableImpactsAnalyzer(SymbolicRegressionVariableImpactsAnalyzer original, Cloner cloner)
    150             : base(original, cloner)
    151         {
    152             impactsCalculator = new SymbolicRegressionSolutionImpactValuesCalculator();
    153             simplifier = new SymbolicDataAnalysisExpressionTreeSimplifier();
    154         }
    155 
    156         public override IDeepCloneable Clone(Cloner cloner)
    157         {
    158             return new SymbolicRegressionVariableImpactsAnalyzer(this, cloner);
    159         }
    160 
    161         public override IOperation Apply()
    162         {
    163             #region Update counter & update interval
    164             UpdateCounter++;
    165             if (UpdateCounter != UpdateInterval)
    166             {
    167                 return base.Apply();
    168             }
    169             UpdateCounter = 0;
    170             #endregion
    171             var results = ResultCollectionParameter.ActualValue;
    172             int maxGen = MaximumGenerationsParameter.ActualValue.Value;
    173             int gen = ((IntValue)results["Generations"].Value).Value;
    174             int lastGen = LastGenerationsParameter.Value.Value;
    175 
    176             if (gen < maxGen - lastGen)
    177                 return base.Apply();
    178 
    179             var trees = SymbolicExpressionTree.ToArray();
    180             var qualities = QualityParameter.ActualValue.ToArray();
    181 
    182             Array.Sort(qualities, trees);
    183             Array.Reverse(qualities);
    184             Array.Reverse(trees);
    185 
    186             var interpreter = SymbolicDataAnalysisTreeInterpreterParameter.ActualValue;
    187             var problemData = ProblemDataParameter.ActualValue;
    188             var applyLinearScaling = ApplyLinearScalingParameter.ActualValue.Value;
    189             var maxCOIterations = MaxCOIterationsParameter.Value.Value; // fixed value parameter => Value
    190             var estimationLimits = EstimationLimitsParameter.ActualValue; // lookup parameter => ActualValue
    191             var percentageBest = PercentageBestParameter.Value.Value;
    192             var optimizeConstants = OptimizeConstantsParameter.Value.Value;
    193             var pruneTrees = PruneTreesParameter.Value.Value;
    194 
    195             var allowedInputVariables = problemData.AllowedInputVariables.ToList();
    196             var variableImpacts = allowedInputVariables.ToDictionary(x => x, x => 0.0);
    197             DataTable datatable;
    198             if (VariableImpactsParameter.ActualValue == null)
    199             {
    200                 datatable = new DataTable("Variable impacts", "Average impact of variables over the population");
    201                 datatable.VisualProperties.XAxisTitle = "Generation";
    202                 datatable.VisualProperties.YAxisTitle = "Average variable impact";
    203                 VariableImpactsParameter.ActualValue = datatable;
    204                 results.Add(new Result("Average variable impacts", "The relative variable relevance calculated as the average relative variable frequency over the whole run.", new DataTable()));
    205 
    206                 foreach (var v in allowedInputVariables)
    207                 {
    208                     datatable.Rows.Add(new DataRow(v) { VisualProperties = { StartIndexZero = true } });
    209                 }
    210                 VariableImpactsParameter.ActualValue = datatable;
    211             }
    212             datatable = VariableImpactsParameter.ActualValue;
    213             int nTrees = (int)Math.Round(trees.Length * percentageBest);
    214             var bestTrees = trees.Take(nTrees).Select(x => (ISymbolicExpressionTree)x.Clone()).ToList();
    215 
    216             if (optimizeConstants)
    217             {
    218                 foreach (var tree in bestTrees)
    219                     SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, tree, problemData, problemData.TrainingIndices, applyLinearScaling, maxCOIterations, estimationLimits.Upper, estimationLimits.Lower);
    220             }
    221             var simplifiedTrees = bestTrees.Select(x => simplifier.Simplify(x)).ToList();
    222             var variableCounts = problemData.AllowedInputVariables.ToDictionary(x => x, x => simplifiedTrees.Count(t => t.IterateNodesPrefix().Any(n => n is VariableTreeNode && ((VariableTreeNode)n).VariableName == x)));
    223             for (int i = 0; i < simplifiedTrees.Count; ++i)
    224             {
    225                 var simplifiedTree = simplifiedTrees[i];
    226 
    227                 if (pruneTrees)
    228                     simplifiedTree = SymbolicRegressionPruningOperator.Prune(simplifiedTree, impactsCalculator, interpreter, problemData, estimationLimits, problemData.TrainingIndices);
    229 
    230                 var quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(interpreter, simplifiedTree, estimationLimits.Lower, estimationLimits.Upper, problemData, problemData.TrainingIndices, applyLinearScaling);
    231 
    232                 var model = new SymbolicRegressionModel(simplifiedTree, interpreter, estimationLimits.Lower, estimationLimits.Upper);
    233                 var variables = simplifiedTree.IterateNodesPrefix().Where(x => x is VariableTreeNode).GroupBy(x => ((VariableTreeNode)x).VariableName);
    234 
    235                 foreach (var g in variables)
    236                 {
    237                     var avgImpact = g.Average(x => impactsCalculator.CalculateImpactValue(model, x, problemData, problemData.TrainingIndices, quality));
    238                     if (double.IsNaN(avgImpact))
    239                         avgImpact = 0;
    240                     variableImpacts[g.Key] += avgImpact;
    241                 }
    242             }
    243 
    244             foreach (var pair in variableImpacts)
    245             {
    246                 var v = variableCounts[pair.Key] > 0 ? pair.Value / variableCounts[pair.Key] : 0;
    247                 datatable.Rows[pair.Key].Values.Add(v);
    248             }
    249             results["Average variable impacts"].Value = datatable;
    250             return base.Apply();
    251         }
    252     }
     35
     36namespace HeuristicLab.VariableInteractionNetworks {
     37  [Item("SymbolicRegressionVariableImpactsAnalyzer", "An analyzer which calculates variable impacts based on the average node impacts from the tree")]
     38  [StorableClass]
     39  public class SymbolicRegressionVariableImpactsAnalyzer : SymbolicDataAnalysisAnalyzer {
     40    private const string UpdateCounterParameterName = "UpdateCounter";
     41    private const string UpdateIntervalParameterName = "UpdateInterval";
     42    public const string QualityParameterName = "Quality";
     43    private const string SymbolicDataAnalysisTreeInterpreterParameterName = "SymbolicExpressionTreeInterpreter";
     44    private const string ProblemDataParameterName = "ProblemData";
     45    private const string ApplyLinearScalingParameterName = "ApplyLinearScaling";
     46    private const string MaxCOIterationsParameterName = "MaxCOIterations";
     47    private const string EstimationLimitsParameterName = "EstimationLimits";
     48    private const string EvaluatorParameterName = "Evaluator";
     49    private const string VariableImpactsParameterName = "AverageVariableImpacts";
     50    private const string PercentageBestParameterName = "PercentageBest";
     51    private const string LastGenerationsParameterName = "LastGenerations";
     52    private const string MaximumGenerationsParameterName = "MaximumGenerations";
     53    private const string OptimizeConstantsParameterName = "OptimizeConstants";
     54    private const string PruneTreesParameterName = "PruneTrees";
     55
     56    private SymbolicDataAnalysisExpressionTreeSimplifier simplifier;
     57    private SymbolicRegressionSolutionImpactValuesCalculator impactsCalculator;
     58
     59    #region parameters
     60    public ValueParameter<IntValue> UpdateCounterParameter {
     61      get { return (ValueParameter<IntValue>)Parameters[UpdateCounterParameterName]; }
     62    }
     63    public ValueParameter<IntValue> UpdateIntervalParameter {
     64      get { return (ValueParameter<IntValue>)Parameters[UpdateIntervalParameterName]; }
     65    }
     66    public IScopeTreeLookupParameter<DoubleValue> QualityParameter {
     67      get { return (IScopeTreeLookupParameter<DoubleValue>)Parameters[QualityParameterName]; }
     68    }
     69    public ILookupParameter<ISymbolicDataAnalysisExpressionTreeInterpreter> SymbolicDataAnalysisTreeInterpreterParameter {
     70      get { return (ILookupParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>)Parameters[SymbolicDataAnalysisTreeInterpreterParameterName]; }
     71    }
     72    public ILookupParameter<IRegressionProblemData> ProblemDataParameter {
     73      get { return (ILookupParameter<IRegressionProblemData>)Parameters[ProblemDataParameterName]; }
     74    }
     75    public ILookupParameter<BoolValue> ApplyLinearScalingParameter {
     76      get { return (ILookupParameter<BoolValue>)Parameters[ApplyLinearScalingParameterName]; }
     77    }
     78    public IFixedValueParameter<IntValue> MaxCOIterationsParameter {
     79      get { return (IFixedValueParameter<IntValue>)Parameters[MaxCOIterationsParameterName]; }
     80    }
     81    public ILookupParameter<DoubleLimit> EstimationLimitsParameter {
     82      get { return (ILookupParameter<DoubleLimit>)Parameters[EstimationLimitsParameterName]; }
     83    }
     84    public ILookupParameter<ISymbolicRegressionSingleObjectiveEvaluator> EvaluatorParameter {
     85      get { return (ILookupParameter<ISymbolicRegressionSingleObjectiveEvaluator>)Parameters[EvaluatorParameterName]; }
     86    }
     87    public ILookupParameter<DataTable> VariableImpactsParameter {
     88      get { return (ILookupParameter<DataTable>)Parameters[VariableImpactsParameterName]; }
     89    }
     90    public IFixedValueParameter<PercentValue> PercentageBestParameter {
     91      get { return (IFixedValueParameter<PercentValue>)Parameters[PercentageBestParameterName]; }
     92    }
     93    public IFixedValueParameter<IntValue> LastGenerationsParameter {
     94      get { return (IFixedValueParameter<IntValue>)Parameters[LastGenerationsParameterName]; }
     95    }
     96    public IFixedValueParameter<BoolValue> OptimizeConstantsParameter {
     97      get { return (IFixedValueParameter<BoolValue>)Parameters[OptimizeConstantsParameterName]; }
     98    }
     99    public IFixedValueParameter<BoolValue> PruneTreesParameter {
     100      get { return (IFixedValueParameter<BoolValue>)Parameters[PruneTreesParameterName]; }
     101    }
     102    private ILookupParameter<IntValue> MaximumGenerationsParameter {
     103      get { return (ILookupParameter<IntValue>)Parameters[MaximumGenerationsParameterName]; }
     104    }
     105    #endregion
     106
     107    #region parameter properties
     108    public int UpdateCounter {
     109      get { return UpdateCounterParameter.Value.Value; }
     110      set { UpdateCounterParameter.Value.Value = value; }
     111    }
     112    public int UpdateInterval {
     113      get { return UpdateIntervalParameter.Value.Value; }
     114      set { UpdateIntervalParameter.Value.Value = value; }
     115    }
     116    #endregion
     117
     118    public SymbolicRegressionVariableImpactsAnalyzer() {
     119      #region add parameters
     120      Parameters.Add(new ValueParameter<IntValue>(UpdateCounterParameterName, new IntValue(0)));
     121      Parameters.Add(new ValueParameter<IntValue>(UpdateIntervalParameterName, new IntValue(1)));
     122      Parameters.Add(new LookupParameter<IRegressionProblemData>(ProblemDataParameterName));
     123      Parameters.Add(new LookupParameter<ISymbolicDataAnalysisExpressionTreeInterpreter>(SymbolicDataAnalysisTreeInterpreterParameterName));
     124      Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>(QualityParameterName, "The individual qualities."));
     125      Parameters.Add(new LookupParameter<BoolValue>(ApplyLinearScalingParameterName));
     126      Parameters.Add(new LookupParameter<DoubleLimit>(EstimationLimitsParameterName));
     127      Parameters.Add(new FixedValueParameter<IntValue>(MaxCOIterationsParameterName, new IntValue(3)));
     128      Parameters.Add(new LookupParameter<DataTable>(VariableImpactsParameterName, "The relative variable relevance calculated as the average relative variable frequency over the whole run."));
     129      Parameters.Add(new FixedValueParameter<PercentValue>(PercentageBestParameterName, new PercentValue(100)));
     130      Parameters.Add(new FixedValueParameter<IntValue>(LastGenerationsParameterName, new IntValue(10)));
     131      Parameters.Add(new FixedValueParameter<BoolValue>(OptimizeConstantsParameterName, new BoolValue(false)));
     132      Parameters.Add(new FixedValueParameter<BoolValue>(PruneTreesParameterName, new BoolValue(false)));
     133      Parameters.Add(new LookupParameter<IntValue>(MaximumGenerationsParameterName, "The maximum number of generations which should be processed."));
     134      Parameters.Add(new LookupParameter<ISymbolicRegressionSingleObjectiveEvaluator>(EvaluatorParameterName));
     135      #endregion
     136
     137      impactsCalculator = new SymbolicRegressionSolutionImpactValuesCalculator();
     138      simplifier = new SymbolicDataAnalysisExpressionTreeSimplifier();
     139    }
     140
     141    [StorableConstructor]
     142    protected SymbolicRegressionVariableImpactsAnalyzer(bool deserializing) : base(deserializing) { }
     143
     144    [StorableHook(HookType.AfterDeserialization)]
     145    private void AfterDeserialization() {
     146      impactsCalculator = new SymbolicRegressionSolutionImpactValuesCalculator();
     147      simplifier = new SymbolicDataAnalysisExpressionTreeSimplifier();
     148
     149      if (!Parameters.ContainsKey(EvaluatorParameterName))
     150        Parameters.Add(new LookupParameter<ISymbolicRegressionSingleObjectiveEvaluator>(EvaluatorParameterName));
     151    }
     152
     153    protected SymbolicRegressionVariableImpactsAnalyzer(SymbolicRegressionVariableImpactsAnalyzer original, Cloner cloner)
     154        : base(original, cloner) {
     155      impactsCalculator = new SymbolicRegressionSolutionImpactValuesCalculator();
     156      simplifier = new SymbolicDataAnalysisExpressionTreeSimplifier();
     157    }
     158
     159    public override IDeepCloneable Clone(Cloner cloner) {
     160      return new SymbolicRegressionVariableImpactsAnalyzer(this, cloner);
     161    }
     162
     163    public override IOperation Apply() {
     164      #region Update counter & update interval
     165      UpdateCounter++;
     166      if (UpdateCounter != UpdateInterval) {
     167        return base.Apply();
     168      }
     169      UpdateCounter = 0;
     170      #endregion
     171      var results = ResultCollectionParameter.ActualValue;
     172      int maxGen = MaximumGenerationsParameter.ActualValue.Value;
     173      int gen = ((IntValue)results["Generations"].Value).Value;
     174      int lastGen = LastGenerationsParameter.Value.Value;
     175
     176      if (lastGen > 0 && gen < maxGen - lastGen)
     177        return base.Apply();
     178
     179      var trees = SymbolicExpressionTree.ToArray();
     180      var qualities = QualityParameter.ActualValue.ToArray();
     181
     182      Array.Sort(qualities, trees);
     183      Array.Reverse(qualities);
     184      Array.Reverse(trees);
     185
     186      var interpreter = SymbolicDataAnalysisTreeInterpreterParameter.ActualValue;
     187      var problemData = ProblemDataParameter.ActualValue;
     188      var applyLinearScaling = ApplyLinearScalingParameter.ActualValue.Value;
     189      var constantOptimizationIterations = MaxCOIterationsParameter.Value.Value; // fixed value parameter => Value
     190      var estimationLimits = EstimationLimitsParameter.ActualValue; // lookup parameter => ActualValue
     191      var percentageBest = PercentageBestParameter.Value.Value;
     192      var optimizeConstants = OptimizeConstantsParameter.Value.Value;
     193      var pruneTrees = PruneTreesParameter.Value.Value;
     194
     195      var allowedInputVariables = problemData.AllowedInputVariables.ToList();
     196      DataTable dataTable;
     197      if (VariableImpactsParameter.ActualValue == null) {
     198        dataTable = new DataTable("Variable impacts", "Average impact of variables over the population");
     199        dataTable.VisualProperties.XAxisTitle = "Generation";
     200        dataTable.VisualProperties.YAxisTitle = "Average variable impact";
     201        VariableImpactsParameter.ActualValue = dataTable;
     202        results.Add(new Result("Average variable impacts", "The relative variable relevance calculated as the average relative variable frequency over the whole run.", new DataTable()));
     203
     204        foreach (var v in allowedInputVariables) {
     205          dataTable.Rows.Add(new DataRow(v) { VisualProperties = { StartIndexZero = true } });
     206        }
     207        VariableImpactsParameter.ActualValue = dataTable;
     208      }
     209      dataTable = VariableImpactsParameter.ActualValue;
     210      int nTrees = (int)Math.Round(trees.Length * percentageBest);
     211      var bestTrees = trees.Take(nTrees).Select(x => (ISymbolicExpressionTree)x.Clone()).ToList();
     212      // simplify trees before doing anything else
     213      var simplifiedTrees = bestTrees.Select(x => simplifier.Simplify(x)).ToList();
     214
     215      if (optimizeConstants) {
     216        for (int i = 0; i < simplifiedTrees.Count; ++i) {
     217          qualities[i].Value = SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, simplifiedTrees[i], problemData, problemData.TrainingIndices, applyLinearScaling, constantOptimizationIterations, estimationLimits.Upper, estimationLimits.Lower);
     218        }
     219      }
     220
     221      if (pruneTrees) {
     222        for (int i = 0; i < simplifiedTrees.Count; ++i) {
     223          simplifiedTrees[i] = SymbolicRegressionPruningOperator.Prune(simplifiedTrees[i], impactsCalculator, interpreter, problemData, estimationLimits, problemData.TrainingIndices);
     224        }
     225      }
     226      // map each variable to a list of indices of the trees that contain it
     227      var variablesToTreeIndices = allowedInputVariables.ToDictionary(x => x, x => Enumerable.Range(0, simplifiedTrees.Count).Where(i => ContainsVariable(simplifiedTrees[i], x)).ToList());
     228
     229      foreach (var mapping in variablesToTreeIndices) {
     230        var variableName = mapping.Key;
     231        var median = problemData.Dataset.GetDoubleValues(variableName, problemData.TrainingIndices).Median();
     232        var ds = new ModifiableDataset(problemData.Dataset.DoubleVariables, problemData.Dataset.DoubleVariables.Select(x => problemData.Dataset.GetReadOnlyDoubleValues(x).ToList()));
     233        foreach (var i in problemData.TrainingIndices) {
     234          ds.SetVariableValue(median, variableName, i);
     235        }
     236        var pd = new RegressionProblemData(ds, allowedInputVariables, problemData.TargetVariable);
     237        pd.TrainingPartition.Start = problemData.TrainingPartition.Start;
     238        pd.TrainingPartition.End = problemData.TrainingPartition.End;
     239        pd.TestPartition.Start = problemData.TestPartition.Start;
     240        pd.TestPartition.End = problemData.TestPartition.End;
     241
     242        var indices = mapping.Value;
     243        var averageImpact = 0d;
     244        for (int i = 0; i < indices.Count; ++i) {
     245          var originalQuality = qualities[i].Value;
     246          double newQuality;
     247          if (optimizeConstants) {
     248            newQuality = SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, simplifiedTrees[i], pd, problemData.TrainingIndices, applyLinearScaling, constantOptimizationIterations, estimationLimits.Upper, estimationLimits.Lower);
     249          } else {
     250            var evaluator = EvaluatorParameter.ActualValue;
     251            newQuality = evaluator.Evaluate(this.ExecutionContext, simplifiedTrees[i], pd, pd.TrainingIndices);
     252          }
     253          averageImpact += originalQuality - newQuality; // impact calculated this way may be negative
     254        }
     255        averageImpact /= indices.Count;
     256        dataTable.Rows[variableName].Values.Add(averageImpact);
     257      }
     258
     259      results["Average variable impacts"].Value = dataTable;
     260      return base.Apply();
     261    }
     262
     263    private static bool ContainsVariable(ISymbolicExpressionTree tree, string variableName) {
     264      return tree.IterateNodesPrefix().OfType<VariableTreeNode>().Any(x => x.VariableName == variableName);
     265    }
     266  }
    253267}
Note: See TracChangeset for help on using the changeset viewer.