Changeset 16536


Ignore:
Timestamp:
01/17/19 14:36:59 (2 years ago)
Author:
chaider
Message:

#2971 merged DataAnalysis.Views from trunk to branch

Location:
branches/2971_named_intervals
Files:
6 edited

Legend:

Unmodified
Added
Removed
  • branches/2971_named_intervals/HeuristicLab.Problems.DataAnalysis

  • branches/2971_named_intervals/HeuristicLab.Problems.DataAnalysis.Views/3.4/HeuristicLab.Problems.DataAnalysis.Views-3.4.csproj

    r15638 r16536  
    103103      <Private>False</Private>
    104104    </Reference>
     105    <Reference Include="HeuristicLab.Algorithms.DataAnalysis-3.4">
     106      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Algorithms.DataAnalysis-3.4.dll</HintPath>
     107    </Reference>
     108    <Reference Include="HeuristicLab.Analysis-3.3">
     109      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Analysis-3.3.dll</HintPath>
     110    </Reference>
     111    <Reference Include="HeuristicLab.Analysis.Views-3.3">
     112      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Analysis.Views-3.3.dll</HintPath>
     113    </Reference>
     114    <Reference Include="HeuristicLab.Collections-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     115      <SpecificVersion>False</SpecificVersion>
     116      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Collections-3.3.dll</HintPath>
     117    </Reference>
     118    <Reference Include="HeuristicLab.Common-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     119      <SpecificVersion>False</SpecificVersion>
     120      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Common-3.3.dll</HintPath>
     121    </Reference>
     122    <Reference Include="HeuristicLab.Common.Resources-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     123      <SpecificVersion>False</SpecificVersion>
     124      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Common.Resources-3.3.dll</HintPath>
     125    </Reference>
     126    <Reference Include="HeuristicLab.Core-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     127      <SpecificVersion>False</SpecificVersion>
     128      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Core-3.3.dll</HintPath>
     129    </Reference>
     130    <Reference Include="HeuristicLab.Core.Views-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     131      <SpecificVersion>False</SpecificVersion>
     132      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Core.Views-3.3.dll</HintPath>
     133    </Reference>
     134    <Reference Include="HeuristicLab.Data-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     135      <SpecificVersion>False</SpecificVersion>
     136      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Data-3.3.dll</HintPath>
     137    </Reference>
     138    <Reference Include="HeuristicLab.MainForm-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     139      <SpecificVersion>False</SpecificVersion>
     140      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.MainForm-3.3.dll</HintPath>
     141    </Reference>
     142    <Reference Include="HeuristicLab.MainForm.WindowsForms-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     143      <SpecificVersion>False</SpecificVersion>
     144      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.MainForm.WindowsForms-3.3.dll</HintPath>
     145    </Reference>
     146    <Reference Include="HeuristicLab.Optimization-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     147      <SpecificVersion>False</SpecificVersion>
     148      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Optimization-3.3.dll</HintPath>
     149    </Reference>
     150    <Reference Include="HeuristicLab.Optimization.Views-3.3">
     151      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Optimization.Views-3.3.dll</HintPath>
     152    </Reference>
     153    <Reference Include="HeuristicLab.Optimizer-3.3">
     154      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Optimizer-3.3.dll</HintPath>
     155    </Reference>
     156    <Reference Include="HeuristicLab.Parameters.Views-3.3">
     157      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Parameters.Views-3.3.dll</HintPath>
     158    </Reference>
     159    <Reference Include="HeuristicLab.Persistence-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     160      <SpecificVersion>False</SpecificVersion>
     161      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Persistence-3.3.dll</HintPath>
     162    </Reference>
     163    <Reference Include="HeuristicLab.PluginInfrastructure-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">
     164      <SpecificVersion>False</SpecificVersion>
     165      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.PluginInfrastructure-3.3.dll</HintPath>
     166    </Reference>
     167    <Reference Include="HeuristicLab.Problems.DataAnalysis.Symbolic-3.4">
     168      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Problems.DataAnalysis.Symbolic-3.4.dll</HintPath>
     169    </Reference>
     170    <Reference Include="HeuristicLab.Problems.DataAnalysis.Symbolic.Regression-3.4">
     171      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Problems.DataAnalysis.Symbolic.Regression-3.4.dll</HintPath>
     172    </Reference>
     173    <Reference Include="HeuristicLab.Visualization.ChartControlsExtensions-3.3">
     174      <HintPath>..\..\..\..\trunk\bin\HeuristicLab.Visualization.ChartControlsExtensions-3.3.dll</HintPath>
     175    </Reference>
    105176    <Reference Include="System" />
    106177    <Reference Include="System.Core">
     
    472543  </ItemGroup>
    473544  <ItemGroup>
    474     <ProjectReference Include="..\..\HeuristicLab.Algorithms.DataAnalysis\3.4\HeuristicLab.Algorithms.DataAnalysis-3.4.csproj">
    475       <Project>{2E782078-FA81-4B70-B56F-74CE38DAC6C8}</Project>
    476       <Name>HeuristicLab.Algorithms.DataAnalysis-3.4</Name>
    477       <Private>False</Private>
    478     </ProjectReference>
    479     <ProjectReference Include="..\..\HeuristicLab.Analysis.Views\3.3\HeuristicLab.Analysis.Views-3.3.csproj">
    480       <Project>{76945D76-CA61-4147-9DC2-0ACDCDDF87F9}</Project>
    481       <Name>HeuristicLab.Analysis.Views-3.3</Name>
    482       <Private>False</Private>
    483     </ProjectReference>
    484     <ProjectReference Include="..\..\HeuristicLab.Analysis\3.3\HeuristicLab.Analysis-3.3.csproj">
    485       <Project>{887425B4-4348-49ED-A457-B7D2C26DDBF9}</Project>
    486       <Name>HeuristicLab.Analysis-3.3</Name>
    487       <Private>False</Private>
    488     </ProjectReference>
    489     <ProjectReference Include="..\..\HeuristicLab.Collections\3.3\HeuristicLab.Collections-3.3.csproj">
    490       <Project>{958B43BC-CC5C-4FA2-8628-2B3B01D890B6}</Project>
    491       <Name>HeuristicLab.Collections-3.3</Name>
    492       <Private>False</Private>
    493     </ProjectReference>
    494     <ProjectReference Include="..\..\HeuristicLab.Common.Resources\3.3\HeuristicLab.Common.Resources-3.3.csproj">
    495       <Project>{0E27A536-1C4A-4624-A65E-DC4F4F23E3E1}</Project>
    496       <Name>HeuristicLab.Common.Resources-3.3</Name>
    497       <Private>False</Private>
    498     </ProjectReference>
    499     <ProjectReference Include="..\..\HeuristicLab.Common\3.3\HeuristicLab.Common-3.3.csproj">
    500       <Project>{A9AD58B9-3EF9-4CC1-97E5-8D909039FF5C}</Project>
    501       <Name>HeuristicLab.Common-3.3</Name>
    502       <Private>False</Private>
    503     </ProjectReference>
    504     <ProjectReference Include="..\..\HeuristicLab.Core.Views\3.3\HeuristicLab.Core.Views-3.3.csproj">
    505       <Project>{E226881D-315F-423D-B419-A766FE0D8685}</Project>
    506       <Name>HeuristicLab.Core.Views-3.3</Name>
    507       <Private>False</Private>
    508     </ProjectReference>
    509     <ProjectReference Include="..\..\HeuristicLab.Core\3.3\HeuristicLab.Core-3.3.csproj">
    510       <Project>{C36BD924-A541-4A00-AFA8-41701378DDC5}</Project>
    511       <Name>HeuristicLab.Core-3.3</Name>
    512       <Private>False</Private>
    513     </ProjectReference>
    514545    <ProjectReference Include="..\..\HeuristicLab.Data.Views\3.3\HeuristicLab.Data.Views-3.3.csproj">
    515546      <Project>{72104A0B-90E7-42F3-9ABE-9BBBADD4B943}</Project>
     
    517548      <Private>False</Private>
    518549    </ProjectReference>
    519     <ProjectReference Include="..\..\HeuristicLab.Data\3.3\HeuristicLab.Data-3.3.csproj">
    520       <Project>{BBAB9DF5-5EF3-4BA8-ADE9-B36E82114937}</Project>
    521       <Name>HeuristicLab.Data-3.3</Name>
    522       <Private>False</Private>
    523     </ProjectReference>
    524     <ProjectReference Include="..\..\HeuristicLab.MainForm.WindowsForms\3.3\HeuristicLab.MainForm.WindowsForms-3.3.csproj">
    525       <Project>{AB687BBE-1BFE-476B-906D-44237135431D}</Project>
    526       <Name>HeuristicLab.MainForm.WindowsForms-3.3</Name>
    527       <Private>False</Private>
    528     </ProjectReference>
    529     <ProjectReference Include="..\..\HeuristicLab.MainForm\3.3\HeuristicLab.MainForm-3.3.csproj">
    530       <Project>{3BD61258-31DA-4B09-89C0-4F71FEF5F05A}</Project>
    531       <Name>HeuristicLab.MainForm-3.3</Name>
    532       <Private>False</Private>
    533     </ProjectReference>
    534     <ProjectReference Include="..\..\HeuristicLab.Optimization.Views\3.3\HeuristicLab.Optimization.Views-3.3.csproj">
    535       <Project>{662B4B15-8F4D-4AE5-B3EB-D91C215F5AF2}</Project>
    536       <Name>HeuristicLab.Optimization.Views-3.3</Name>
    537       <Private>False</Private>
    538     </ProjectReference>
    539     <ProjectReference Include="..\..\HeuristicLab.Optimization\3.3\HeuristicLab.Optimization-3.3.csproj">
    540       <Project>{14AB8D24-25BC-400C-A846-4627AA945192}</Project>
    541       <Name>HeuristicLab.Optimization-3.3</Name>
    542       <Private>False</Private>
    543     </ProjectReference>
    544     <ProjectReference Include="..\..\HeuristicLab.Optimizer\3.3\HeuristicLab.Optimizer-3.3.csproj">
    545       <Project>{C664305E-497C-4533-A140-967DEDB05C19}</Project>
    546       <Name>HeuristicLab.Optimizer-3.3</Name>
    547       <Private>False</Private>
    548     </ProjectReference>
    549     <ProjectReference Include="..\..\HeuristicLab.Parameters.Views\3.3\HeuristicLab.Parameters.Views-3.3.csproj">
    550       <Project>{AE5B1CE5-9862-4D6F-A700-D72CD9AEA295}</Project>
    551       <Name>HeuristicLab.Parameters.Views-3.3</Name>
    552       <Private>False</Private>
    553     </ProjectReference>
    554     <ProjectReference Include="..\..\HeuristicLab.Persistence\3.3\HeuristicLab.Persistence-3.3.csproj">
    555       <Project>{102BC7D3-0EF9-439C-8F6D-96FF0FDB8E1B}</Project>
    556       <Name>HeuristicLab.Persistence-3.3</Name>
    557       <Private>False</Private>
    558     </ProjectReference>
    559     <ProjectReference Include="..\..\HeuristicLab.PluginInfrastructure\3.3\HeuristicLab.PluginInfrastructure-3.3.csproj">
    560       <Project>{94186A6A-5176-4402-AE83-886557B53CCA}</Project>
    561       <Name>HeuristicLab.PluginInfrastructure-3.3</Name>
    562       <Private>False</Private>
    563     </ProjectReference>
    564     <ProjectReference Include="..\..\HeuristicLab.Problems.DataAnalysis.Symbolic.Regression\3.4\HeuristicLab.Problems.DataAnalysis.Symbolic.Regression-3.4.csproj">
    565       <Project>{5AC82412-911B-4FA2-A013-EDC5E3F3FCC2}</Project>
    566       <Name>HeuristicLab.Problems.DataAnalysis.Symbolic.Regression-3.4</Name>
    567       <Private>False</Private>
    568     </ProjectReference>
    569     <ProjectReference Include="..\..\HeuristicLab.Problems.DataAnalysis.Symbolic\3.4\HeuristicLab.Problems.DataAnalysis.Symbolic-3.4.csproj">
    570       <Project>{3D28463F-EC96-4D82-AFEE-38BE91A0CA00}</Project>
    571       <Name>HeuristicLab.Problems.DataAnalysis.Symbolic-3.4</Name>
    572       <Private>False</Private>
    573     </ProjectReference>
    574550    <ProjectReference Include="..\..\HeuristicLab.Problems.DataAnalysis\3.4\HeuristicLab.Problems.DataAnalysis-3.4.csproj">
    575551      <Project>{DF87C13E-A889-46FF-8153-66DCAA8C5674}</Project>
    576552      <Name>HeuristicLab.Problems.DataAnalysis-3.4</Name>
    577       <Private>False</Private>
    578     </ProjectReference>
    579     <ProjectReference Include="..\..\HeuristicLab.Visualization.ChartControlsExtensions\3.3\HeuristicLab.Visualization.ChartControlsExtensions-3.3.csproj">
    580       <Project>{315BDA09-3F4F-49B3-9790-B37CFC1C5750}</Project>
    581       <Name>HeuristicLab.Visualization.ChartControlsExtensions-3.3</Name>
    582553      <Private>False</Private>
    583554    </ProjectReference>
  • branches/2971_named_intervals/HeuristicLab.Problems.DataAnalysis/3.4

    • Property svn:mergeinfo set to (toggle deleted branches)
      /branches/2839_HiveProjectManagement/HeuristicLab.Problems.DataAnalysis/3.4mergedeligible
      /branches/2915-AbsoluteSymbol/HeuristicLab.Problems.DataAnalysis/3.4mergedeligible
      /branches/2947_ConfigurableIndexedDataTable/HeuristicLab.Problems.DataAnalysis/3.4mergedeligible
      /branches/2965_CancelablePersistence/HeuristicLab.Problems.DataAnalysis/3.4mergedeligible
      /stable/HeuristicLab.Problems.DataAnalysis/3.4mergedeligible
      /trunk/HeuristicLab.Problems.DataAnalysis/3.4mergedeligible
      /branches/2892_LR-prediction-intervals/HeuristicLab.Problems.DataAnalysis/3.415743-16388
      /branches/2904_CalculateImpacts/3.415808-16421
      /branches/2966_interval_calculation/HeuristicLab.Problems.DataAnalysis/3.416320-16406
      /branches/Async/HeuristicLab.Problems.DataAnalysis/3.413329-15286
      /branches/Classification-Extensions/HeuristicLab.Problems.DataAnalysis/3.411606-11761
      /branches/ClassificationModelComparison/HeuristicLab.Problems.DataAnalysis/3.49073-13099
      /branches/CloningRefactoring/HeuristicLab.Problems.DataAnalysis/3.44656-4721
      /branches/DataAnalysis Refactoring/HeuristicLab.Problems.DataAnalysis/3.45471-5808
      /branches/DataAnalysis SolutionEnsembles/HeuristicLab.Problems.DataAnalysis/3.45815-6180
      /branches/DataAnalysis/HeuristicLab.Problems.DataAnalysis/3.44220,​4226,​4236-4238,​4389,​4458-4459,​4462,​4464
      /branches/DataAnalysisCSVImport/HeuristicLab.Problems.DataAnalysis/3.48713-8875
      /branches/DataPreprocessing/HeuristicLab.Problems.DataAnalysis/3.410085-11101
      /branches/DatasetFeatureCorrelation/HeuristicLab.Problems.DataAnalysis/3.48035-8538
      /branches/GP.Grammar.Editor/HeuristicLab.Problems.DataAnalysis/3.46284-6795
      /branches/GP.Symbols (TimeLag, Diff, Integral)/HeuristicLab.Problems.DataAnalysis/3.45060
      /branches/HeuristicLab.DatasetRefactor/sources/HeuristicLab.Problems.DataAnalysis/3.411570-12508
      /branches/HeuristicLab.Problems.Orienteering/HeuristicLab.Problems.DataAnalysis/3.411130-12721
      /branches/HeuristicLab.RegressionSolutionGradientView/HeuristicLab.Problems.DataAnalysis/3.413819-14091
      /branches/HeuristicLab.TimeSeries/HeuristicLab.Problems.DataAnalysis/3.47098-8789
      /branches/LogResidualEvaluator/HeuristicLab.Problems.DataAnalysis/3.410202-10483
      /branches/NET40/sources/HeuristicLab.Problems.DataAnalysis/3.45138-5162
      /branches/ParallelEngine/HeuristicLab.Problems.DataAnalysis/3.45175-5192
      /branches/ProblemInstancesRegressionAndClassification/HeuristicLab.Problems.DataAnalysis/3.47570-7810
      /branches/QAPAlgorithms/HeuristicLab.Problems.DataAnalysis/3.46350-6627
      /branches/Restructure trunk solution/HeuristicLab.Problems.DataAnalysis/3.46828
      /branches/SimplifierViewsProgress/HeuristicLab.Problems.DataAnalysis/3.415318-15370
      /branches/SpectralKernelForGaussianProcesses/HeuristicLab.Problems.DataAnalysis/3.410204-10479
      /branches/Trunk/HeuristicLab.Problems.DataAnalysis/3.46829-6865
      /branches/histogram/HeuristicLab.Problems.DataAnalysis/3.45959-6341
      /branches/symbreg-factors-2650/HeuristicLab.Problems.DataAnalysis/3.414232-14825
  • branches/2971_named_intervals/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Classification/ClassificationSolutionVariableImpactsCalculator.cs

    r15871 r16536  
    2323
    2424using System;
     25using System.Collections;
    2526using System.Collections.Generic;
    2627using System.Linq;
     
    3637  [Item("ClassificationSolution Impacts Calculator", "Calculation of the impacts of input variables for any classification solution")]
    3738  public sealed class ClassificationSolutionVariableImpactsCalculator : ParameterizedNamedItem {
     39    #region Parameters/Properties
    3840    public enum ReplacementMethodEnum {
    3941      Median,
     
    5456
    5557    private const string ReplacementParameterName = "Replacement Method";
     58    private const string FactorReplacementParameterName = "Factor Replacement Method";
    5659    private const string DataPartitionParameterName = "DataPartition";
    5760
    5861    public IFixedValueParameter<EnumValue<ReplacementMethodEnum>> ReplacementParameter {
    5962      get { return (IFixedValueParameter<EnumValue<ReplacementMethodEnum>>)Parameters[ReplacementParameterName]; }
     63    }
     64    public IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>> FactorReplacementParameter {
     65      get { return (IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>>)Parameters[FactorReplacementParameterName]; }
    6066    }
    6167    public IFixedValueParameter<EnumValue<DataPartitionEnum>> DataPartitionParameter {
     
    6773      set { ReplacementParameter.Value.Value = value; }
    6874    }
     75    public FactorReplacementMethodEnum FactorReplacementMethod {
     76      get { return FactorReplacementParameter.Value.Value; }
     77      set { FactorReplacementParameter.Value.Value = value; }
     78    }
    6979    public DataPartitionEnum DataPartition {
    7080      get { return DataPartitionParameter.Value.Value; }
    7181      set { DataPartitionParameter.Value.Value = value; }
    7282    }
    73 
    74 
     83    #endregion
     84
     85    #region Ctor/Cloner
    7586    [StorableConstructor]
    7687    private ClassificationSolutionVariableImpactsCalculator(bool deserializing) : base(deserializing) { }
    7788    private ClassificationSolutionVariableImpactsCalculator(ClassificationSolutionVariableImpactsCalculator original, Cloner cloner)
    7889      : base(original, cloner) { }
     90    public ClassificationSolutionVariableImpactsCalculator()
     91      : base() {
     92      Parameters.Add(new FixedValueParameter<EnumValue<ReplacementMethodEnum>>(ReplacementParameterName, "The replacement method for variables during impact calculation.", new EnumValue<ReplacementMethodEnum>(ReplacementMethodEnum.Shuffle)));
     93      Parameters.Add(new FixedValueParameter<EnumValue<FactorReplacementMethodEnum>>(FactorReplacementParameterName, "The replacement method for factor variables during impact calculation.", new EnumValue<FactorReplacementMethodEnum>(FactorReplacementMethodEnum.Best)));
     94      Parameters.Add(new FixedValueParameter<EnumValue<DataPartitionEnum>>(DataPartitionParameterName, "The data partition on which the impacts are calculated.", new EnumValue<DataPartitionEnum>(DataPartitionEnum.Training)));
     95    }
     96
    7997    public override IDeepCloneable Clone(Cloner cloner) {
    8098      return new ClassificationSolutionVariableImpactsCalculator(this, cloner);
    8199    }
    82 
    83     public ClassificationSolutionVariableImpactsCalculator()
    84       : base() {
    85       Parameters.Add(new FixedValueParameter<EnumValue<ReplacementMethodEnum>>(ReplacementParameterName, "The replacement method for variables during impact calculation.", new EnumValue<ReplacementMethodEnum>(ReplacementMethodEnum.Median)));
    86       Parameters.Add(new FixedValueParameter<EnumValue<DataPartitionEnum>>(DataPartitionParameterName, "The data partition on which the impacts are calculated.", new EnumValue<DataPartitionEnum>(DataPartitionEnum.Training)));
    87     }
     100    #endregion
    88101
    89102    //mkommend: annoying name clash with static method, open to better naming suggestions
    90103    public IEnumerable<Tuple<string, double>> Calculate(IClassificationSolution solution) {
    91       return CalculateImpacts(solution, DataPartition, ReplacementMethod);
     104      return CalculateImpacts(solution, ReplacementMethod, FactorReplacementMethod, DataPartition);
    92105    }
    93106
    94107    public static IEnumerable<Tuple<string, double>> CalculateImpacts(
    95108      IClassificationSolution solution,
    96       DataPartitionEnum data = DataPartitionEnum.Training,
    97       ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
     109      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
     110      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
     111      DataPartitionEnum dataPartition = DataPartitionEnum.Training) {
     112
     113      IEnumerable<int> rows = GetPartitionRows(dataPartition, solution.ProblemData);
     114      IEnumerable<double> estimatedClassValues = solution.GetEstimatedClassValues(rows);
     115      var model = (IClassificationModel)solution.Model.Clone(); //mkommend: clone of model is necessary, because the thresholds for IDiscriminantClassificationModels are updated
     116
     117      return CalculateImpacts(model, solution.ProblemData, estimatedClassValues, rows, replacementMethod, factorReplacementMethod);
     118    }
     119
     120    public static IEnumerable<Tuple<string, double>> CalculateImpacts(
     121     IClassificationModel model,
     122     IClassificationProblemData problemData,
     123     IEnumerable<double> estimatedClassValues,
     124     IEnumerable<int> rows,
     125     ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
     126     FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
     127
     128      //fholzing: try and catch in case a different dataset is loaded, otherwise statement is neglectable
     129      var missingVariables = model.VariablesUsedForPrediction.Except(problemData.Dataset.VariableNames);
     130      if (missingVariables.Any()) {
     131        throw new InvalidOperationException(string.Format("Can not calculate variable impacts, because the model uses inputs missing in the dataset ({0})", string.Join(", ", missingVariables)));
     132      }
     133      IEnumerable<double> targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
     134      var originalQuality = CalculateQuality(targetValues, estimatedClassValues);
     135
     136      var impacts = new Dictionary<string, double>();
     137      var inputvariables = new HashSet<string>(problemData.AllowedInputVariables.Union(model.VariablesUsedForPrediction));
     138      var modifiableDataset = ((Dataset)(problemData.Dataset).Clone()).ToModifiable();
     139
     140      foreach (var inputVariable in inputvariables) {
     141        impacts[inputVariable] = CalculateImpact(inputVariable, model, problemData, modifiableDataset, rows, replacementMethod, factorReplacementMethod, targetValues, originalQuality);
     142      }
     143
     144      return impacts.Select(i => Tuple.Create(i.Key, i.Value));
     145    }
     146
     147    public static double CalculateImpact(string variableName,
     148      IClassificationModel model,
     149      IClassificationProblemData problemData,
     150      ModifiableDataset modifiableDataset,
     151      IEnumerable<int> rows,
     152      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
     153      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
     154      IEnumerable<double> targetValues = null,
     155      double quality = double.NaN) {
     156
     157      if (!model.VariablesUsedForPrediction.Contains(variableName)) { return 0.0; }
     158      if (!problemData.Dataset.VariableNames.Contains(variableName)) {
     159        throw new InvalidOperationException(string.Format("Can not calculate variable impact, because the model uses inputs missing in the dataset ({0})", variableName));
     160      }
     161
     162      if (targetValues == null) {
     163        targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
     164      }
     165      if (quality == double.NaN) {
     166        quality = CalculateQuality(model.GetEstimatedClassValues(modifiableDataset, rows), targetValues);
     167      }
     168
     169      IList originalValues = null;
     170      IList replacementValues = GetReplacementValues(modifiableDataset, variableName, model, rows, targetValues, out originalValues, replacementMethod, factorReplacementMethod);
     171
     172      double newValue = CalculateQualityForReplacement(model, modifiableDataset, variableName, originalValues, rows, replacementValues, targetValues);
     173      double impact = quality - newValue;
     174
     175      return impact;
     176    }
     177
     178    private static IList GetReplacementValues(ModifiableDataset modifiableDataset,
     179      string variableName,
     180      IClassificationModel model,
     181      IEnumerable<int> rows,
     182      IEnumerable<double> targetValues,
     183      out IList originalValues,
     184      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
    98185      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
    99186
    100       var problemData = solution.ProblemData;
    101       var dataset = problemData.Dataset;
    102       var model = (IClassificationModel)solution.Model.Clone(); //mkommend: clone of model is necessary, because the thresholds for IDiscriminantClassificationModels are updated
    103 
    104       IEnumerable<int> rows;
    105       IEnumerable<double> targetValues;
    106       double originalAccuracy;
    107 
    108       OnlineCalculatorError error;
    109 
    110       switch (data) {
    111         case DataPartitionEnum.All:
    112           rows = problemData.AllIndices;
    113           targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.AllIndices).ToList();
    114           originalAccuracy = OnlineAccuracyCalculator.Calculate(targetValues, solution.EstimatedClassValues, out error);
    115           if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during accuracy calculation.");
    116           break;
    117         case DataPartitionEnum.Training:
    118           rows = problemData.TrainingIndices;
    119           targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TrainingIndices).ToList();
    120           originalAccuracy = OnlineAccuracyCalculator.Calculate(targetValues, solution.EstimatedTrainingClassValues, out error);
    121           if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during accuracy calculation.");
    122           break;
    123         case DataPartitionEnum.Test:
    124           rows = problemData.TestIndices;
    125           targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, problemData.TestIndices).ToList();
    126           originalAccuracy = OnlineAccuracyCalculator.Calculate(targetValues, solution.EstimatedTestClassValues, out error);
    127           if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during accuracy calculation.");
    128           break;
    129         default: throw new ArgumentException(string.Format("DataPartition {0} cannot be handled.", data));
    130       }
    131 
    132       var impacts = new Dictionary<string, double>();
    133       var modifiableDataset = ((Dataset)dataset).ToModifiable();
    134 
    135       var inputvariables = new HashSet<string>(problemData.AllowedInputVariables.Union(solution.Model.VariablesUsedForPrediction));
    136       var allowedInputVariables = dataset.VariableNames.Where(v => inputvariables.Contains(v)).ToList();
    137 
    138       // calculate impacts for double variables
    139       foreach (var inputVariable in allowedInputVariables.Where(problemData.Dataset.VariableHasType<double>)) {
    140         var newEstimates = EvaluateModelWithReplacedVariable(model, inputVariable, modifiableDataset, rows, replacementMethod);
    141         var newAccuracy = OnlineAccuracyCalculator.Calculate(targetValues, newEstimates, out error);
    142         if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during R² calculation with replaced inputs.");
    143 
    144         impacts[inputVariable] = originalAccuracy - newAccuracy;
    145       }
    146 
    147       // calculate impacts for string variables
    148       foreach (var inputVariable in allowedInputVariables.Where(problemData.Dataset.VariableHasType<string>)) {
    149         if (factorReplacementMethod == FactorReplacementMethodEnum.Best) {
    150           // try replacing with all possible values and find the best replacement value
    151           var smallestImpact = double.PositiveInfinity;
    152           foreach (var repl in problemData.Dataset.GetStringValues(inputVariable, rows).Distinct()) {
    153             var newEstimates = EvaluateModelWithReplacedVariable(model, inputVariable, modifiableDataset, rows,
    154               Enumerable.Repeat(repl, dataset.Rows));
    155             var newAccuracy = OnlineAccuracyCalculator.Calculate(targetValues, newEstimates, out error);
    156             if (error != OnlineCalculatorError.None)
    157               throw new InvalidOperationException("Error during accuracy calculation with replaced inputs.");
    158 
    159             var impact = originalAccuracy - newAccuracy;
    160             if (impact < smallestImpact) smallestImpact = impact;
    161           }
    162           impacts[inputVariable] = smallestImpact;
    163         } else {
    164           // for replacement methods shuffle and mode
    165           // calculate impacts for factor variables
    166 
    167           var newEstimates = EvaluateModelWithReplacedVariable(model, inputVariable, modifiableDataset, rows,
    168             factorReplacementMethod);
    169           var newAccuracy = OnlineAccuracyCalculator.Calculate(targetValues, newEstimates, out error);
    170           if (error != OnlineCalculatorError.None)
    171             throw new InvalidOperationException("Error during accuracy calculation with replaced inputs.");
    172 
    173           impacts[inputVariable] = originalAccuracy - newAccuracy;
    174         }
    175       } // foreach
    176       return impacts.OrderByDescending(i => i.Value).Select(i => Tuple.Create(i.Key, i.Value));
    177     }
    178 
    179     private static IEnumerable<double> EvaluateModelWithReplacedVariable(IClassificationModel model, string variable, ModifiableDataset dataset, IEnumerable<int> rows, ReplacementMethodEnum replacement = ReplacementMethodEnum.Median) {
    180       var originalValues = dataset.GetReadOnlyDoubleValues(variable).ToList();
     187      IList replacementValues = null;
     188      if (modifiableDataset.VariableHasType<double>(variableName)) {
     189        originalValues = modifiableDataset.GetReadOnlyDoubleValues(variableName).ToList();
     190        replacementValues = GetReplacementValuesForDouble(modifiableDataset, rows, (List<double>)originalValues, replacementMethod);
     191      } else if (modifiableDataset.VariableHasType<string>(variableName)) {
     192        originalValues = modifiableDataset.GetReadOnlyStringValues(variableName).ToList();
     193        replacementValues = GetReplacementValuesForString(model, modifiableDataset, variableName, rows, (List<string>)originalValues, targetValues, factorReplacementMethod);
     194      } else {
     195        throw new NotSupportedException("Variable not supported");
     196      }
     197
     198      return replacementValues;
     199    }
     200
     201    private static IList GetReplacementValuesForDouble(ModifiableDataset modifiableDataset,
     202      IEnumerable<int> rows,
     203      List<double> originalValues,
     204      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle) {
     205
     206      IRandom random = new FastRandom(31415);
     207      List<double> replacementValues;
    181208      double replacementValue;
    182       List<double> replacementValues;
    183       IRandom rand;
    184 
    185       switch (replacement) {
     209
     210      switch (replacementMethod) {
    186211        case ReplacementMethodEnum.Median:
    187212          replacementValue = rows.Select(r => originalValues[r]).Median();
    188           replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
     213          replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
    189214          break;
    190215        case ReplacementMethodEnum.Average:
    191216          replacementValue = rows.Select(r => originalValues[r]).Average();
    192           replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
     217          replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
    193218          break;
    194219        case ReplacementMethodEnum.Shuffle:
    195220          // new var has same empirical distribution but the relation to y is broken
    196           rand = new FastRandom(31415);
    197221          // prepare a complete column for the dataset
    198           replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
     222          replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
    199223          // shuffle only the selected rows
    200           var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
     224          var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
    201225          int i = 0;
    202226          // update column values
     
    208232          var avg = rows.Select(r => originalValues[r]).Average();
    209233          var stdDev = rows.Select(r => originalValues[r]).StandardDeviation();
    210           rand = new FastRandom(31415);
    211234          // prepare a complete column for the dataset
    212           replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
     235          replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
    213236          // update column values
    214237          foreach (var r in rows) {
    215             replacementValues[r] = NormalDistributedRandom.NextDouble(rand, avg, stdDev);
     238            replacementValues[r] = NormalDistributedRandom.NextDouble(random, avg, stdDev);
    216239          }
    217240          break;
    218241
    219242        default:
    220           throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacement));
    221       }
    222 
    223       return EvaluateModelWithReplacedVariable(model, variable, dataset, rows, replacementValues);
    224     }
    225 
    226     private static IEnumerable<double> EvaluateModelWithReplacedVariable(
    227       IClassificationModel model, string variable, ModifiableDataset dataset,
    228       IEnumerable<int> rows,
    229       FactorReplacementMethodEnum replacement = FactorReplacementMethodEnum.Shuffle) {
    230       var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
    231       List<string> replacementValues;
    232       IRandom rand;
    233 
    234       switch (replacement) {
     243          throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacementMethod));
     244      }
     245
     246      return replacementValues;
     247    }
     248
     249    private static IList GetReplacementValuesForString(IClassificationModel model,
     250      ModifiableDataset modifiableDataset,
     251      string variableName,
     252      IEnumerable<int> rows,
     253      List<string> originalValues,
     254      IEnumerable<double> targetValues,
     255      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Shuffle) {
     256
     257      List<string> replacementValues = null;
     258      IRandom random = new FastRandom(31415);
     259
     260      switch (factorReplacementMethod) {
     261        case FactorReplacementMethodEnum.Best:
     262          // try replacing with all possible values and find the best replacement value
     263          var bestQuality = double.NegativeInfinity;
     264          foreach (var repl in modifiableDataset.GetStringValues(variableName, rows).Distinct()) {
     265            List<string> curReplacementValues = Enumerable.Repeat(repl, modifiableDataset.Rows).ToList();
     266            //fholzing: this result could be used later on (theoretically), but is neglected for better readability/method consistency
     267            var newValue = CalculateQualityForReplacement(model, modifiableDataset, variableName, originalValues, rows, curReplacementValues, targetValues);
     268            var curQuality = newValue;
     269
     270            if (curQuality > bestQuality) {
     271              bestQuality = curQuality;
     272              replacementValues = curReplacementValues;
     273            }
     274          }
     275          break;
    235276        case FactorReplacementMethodEnum.Mode:
    236277          var mostCommonValue = rows.Select(r => originalValues[r])
     
    238279            .OrderByDescending(g => g.Count())
    239280            .First().Key;
    240           replacementValues = Enumerable.Repeat(mostCommonValue, dataset.Rows).ToList();
     281          replacementValues = Enumerable.Repeat(mostCommonValue, modifiableDataset.Rows).ToList();
    241282          break;
    242283        case FactorReplacementMethodEnum.Shuffle:
    243284          // new var has same empirical distribution but the relation to y is broken
    244           rand = new FastRandom(31415);
    245285          // prepare a complete column for the dataset
    246           replacementValues = Enumerable.Repeat(string.Empty, dataset.Rows).ToList();
     286          replacementValues = Enumerable.Repeat(string.Empty, modifiableDataset.Rows).ToList();
    247287          // shuffle only the selected rows
    248           var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
     288          var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
    249289          int i = 0;
    250290          // update column values
     
    254294          break;
    255295        default:
    256           throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", replacement));
    257       }
    258 
    259       return EvaluateModelWithReplacedVariable(model, variable, dataset, rows, replacementValues);
    260     }
    261 
    262     private static IEnumerable<double> EvaluateModelWithReplacedVariable(IClassificationModel model, string variable,
    263       ModifiableDataset dataset, IEnumerable<int> rows, IEnumerable<double> replacementValues) {
    264       var originalValues = dataset.GetReadOnlyDoubleValues(variable).ToList();
    265       dataset.ReplaceVariable(variable, replacementValues.ToList());
    266 
     296          throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", factorReplacementMethod));
     297      }
     298
     299      return replacementValues;
     300    }
     301
     302    private static double CalculateQualityForReplacement(
     303      IClassificationModel model,
     304      ModifiableDataset modifiableDataset,
     305      string variableName,
     306      IList originalValues,
     307      IEnumerable<int> rows,
     308      IList replacementValues,
     309      IEnumerable<double> targetValues) {
     310
     311      modifiableDataset.ReplaceVariable(variableName, replacementValues);
    267312      var discModel = model as IDiscriminantFunctionClassificationModel;
    268313      if (discModel != null) {
    269         var problemData = new ClassificationProblemData(dataset, dataset.VariableNames, model.TargetVariable);
     314        var problemData = new ClassificationProblemData(modifiableDataset, modifiableDataset.VariableNames, model.TargetVariable);
    270315        discModel.RecalculateModelParameters(problemData, rows);
    271316      }
    272317
    273318      //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
    274       var estimates = model.GetEstimatedClassValues(dataset, rows).ToList();
    275       dataset.ReplaceVariable(variable, originalValues);
    276 
    277       return estimates;
    278     }
    279     private static IEnumerable<double> EvaluateModelWithReplacedVariable(IClassificationModel model, string variable,
    280       ModifiableDataset dataset, IEnumerable<int> rows, IEnumerable<string> replacementValues) {
    281       var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
    282       dataset.ReplaceVariable(variable, replacementValues.ToList());
    283 
    284 
    285       var discModel = model as IDiscriminantFunctionClassificationModel;
    286       if (discModel != null) {
    287         var problemData = new ClassificationProblemData(dataset, dataset.VariableNames, model.TargetVariable);
    288         discModel.RecalculateModelParameters(problemData, rows);
    289       }
    290 
    291       //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
    292       var estimates = model.GetEstimatedClassValues(dataset, rows).ToList();
    293       dataset.ReplaceVariable(variable, originalValues);
    294 
    295       return estimates;
     319      var estimates = model.GetEstimatedClassValues(modifiableDataset, rows).ToList();
     320      var ret = CalculateQuality(targetValues, estimates);
     321      modifiableDataset.ReplaceVariable(variableName, originalValues);
     322
     323      return ret;
     324    }
     325
     326    public static double CalculateQuality(IEnumerable<double> targetValues, IEnumerable<double> estimatedClassValues) {
     327      OnlineCalculatorError errorState;
     328      var ret = OnlineAccuracyCalculator.Calculate(targetValues, estimatedClassValues, out errorState);
     329      if (errorState != OnlineCalculatorError.None) { throw new InvalidOperationException("Error during calculation with replaced inputs."); }
     330      return ret;
     331    }
     332
     333    public static IEnumerable<int> GetPartitionRows(DataPartitionEnum dataPartition, IClassificationProblemData problemData) {
     334      IEnumerable<int> rows;
     335
     336      switch (dataPartition) {
     337        case DataPartitionEnum.All:
     338          rows = problemData.AllIndices;
     339          break;
     340        case DataPartitionEnum.Test:
     341          rows = problemData.TestIndices;
     342          break;
     343        case DataPartitionEnum.Training:
     344          rows = problemData.TrainingIndices;
     345          break;
     346        default:
     347          throw new NotSupportedException("DataPartition not supported");
     348      }
     349
     350      return rows;
    296351    }
    297352  }
  • branches/2971_named_intervals/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Regression/RegressionSolutionVariableImpactsCalculator.cs

    r15802 r16536  
    2323
    2424using System;
     25using System.Collections;
    2526using System.Collections.Generic;
    2627using System.Linq;
     
    3637  [Item("RegressionSolution Impacts Calculator", "Calculation of the impacts of input variables for any regression solution")]
    3738  public sealed class RegressionSolutionVariableImpactsCalculator : ParameterizedNamedItem {
     39    #region Parameters/Properties
    3840    public enum ReplacementMethodEnum {
    3941      Median,
     
    5456
    5557    private const string ReplacementParameterName = "Replacement Method";
     58    private const string FactorReplacementParameterName = "Factor Replacement Method";
    5659    private const string DataPartitionParameterName = "DataPartition";
    5760
    5861    public IFixedValueParameter<EnumValue<ReplacementMethodEnum>> ReplacementParameter {
    5962      get { return (IFixedValueParameter<EnumValue<ReplacementMethodEnum>>)Parameters[ReplacementParameterName]; }
     63    }
     64    public IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>> FactorReplacementParameter {
     65      get { return (IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>>)Parameters[FactorReplacementParameterName]; }
    6066    }
    6167    public IFixedValueParameter<EnumValue<DataPartitionEnum>> DataPartitionParameter {
     
    6773      set { ReplacementParameter.Value.Value = value; }
    6874    }
     75    public FactorReplacementMethodEnum FactorReplacementMethod {
     76      get { return FactorReplacementParameter.Value.Value; }
     77      set { FactorReplacementParameter.Value.Value = value; }
     78    }
    6979    public DataPartitionEnum DataPartition {
    7080      get { return DataPartitionParameter.Value.Value; }
    7181      set { DataPartitionParameter.Value.Value = value; }
    7282    }
    73 
    74 
     83    #endregion
     84
     85    #region Ctor/Cloner
    7586    [StorableConstructor]
    7687    private RegressionSolutionVariableImpactsCalculator(bool deserializing) : base(deserializing) { }
    7788    private RegressionSolutionVariableImpactsCalculator(RegressionSolutionVariableImpactsCalculator original, Cloner cloner)
    7889      : base(original, cloner) { }
     90    public RegressionSolutionVariableImpactsCalculator()
     91      : base() {
     92      Parameters.Add(new FixedValueParameter<EnumValue<ReplacementMethodEnum>>(ReplacementParameterName, "The replacement method for variables during impact calculation.", new EnumValue<ReplacementMethodEnum>(ReplacementMethodEnum.Shuffle)));
     93      Parameters.Add(new FixedValueParameter<EnumValue<FactorReplacementMethodEnum>>(FactorReplacementParameterName, "The replacement method for factor variables during impact calculation.", new EnumValue<FactorReplacementMethodEnum>(FactorReplacementMethodEnum.Best)));
     94      Parameters.Add(new FixedValueParameter<EnumValue<DataPartitionEnum>>(DataPartitionParameterName, "The data partition on which the impacts are calculated.", new EnumValue<DataPartitionEnum>(DataPartitionEnum.Training)));
     95    }
     96
    7997    public override IDeepCloneable Clone(Cloner cloner) {
    8098      return new RegressionSolutionVariableImpactsCalculator(this, cloner);
    8199    }
    82 
    83     public RegressionSolutionVariableImpactsCalculator()
    84       : base() {
    85       Parameters.Add(new FixedValueParameter<EnumValue<ReplacementMethodEnum>>(ReplacementParameterName, "The replacement method for variables during impact calculation.", new EnumValue<ReplacementMethodEnum>(ReplacementMethodEnum.Median)));
    86       Parameters.Add(new FixedValueParameter<EnumValue<DataPartitionEnum>>(DataPartitionParameterName, "The data partition on which the impacts are calculated.", new EnumValue<DataPartitionEnum>(DataPartitionEnum.Training)));
    87     }
     100    #endregion
    88101
    89102    //mkommend: annoying name clash with static method, open to better naming suggestions
    90103    public IEnumerable<Tuple<string, double>> Calculate(IRegressionSolution solution) {
    91       return CalculateImpacts(solution, DataPartition, ReplacementMethod);
     104      return CalculateImpacts(solution, ReplacementMethod, FactorReplacementMethod, DataPartition);
    92105    }
    93106
    94107    public static IEnumerable<Tuple<string, double>> CalculateImpacts(
    95108      IRegressionSolution solution,
    96       DataPartitionEnum data = DataPartitionEnum.Training,
    97       ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
     109      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
    98110      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
    99       Func<double, string, bool> progressCallback = null) {
    100 
    101       var problemData = solution.ProblemData;
    102       var dataset = problemData.Dataset;
    103 
    104       IEnumerable<int> rows;
    105       IEnumerable<double> targetValues;
    106       double originalR2 = -1;
    107 
    108       OnlineCalculatorError error;
    109 
    110       switch (data) {
    111         case DataPartitionEnum.All:
    112           rows = solution.ProblemData.AllIndices;
    113           targetValues = problemData.TargetVariableValues.ToList();
    114           originalR2 = OnlinePearsonsRCalculator.Calculate(problemData.TargetVariableValues, solution.EstimatedValues, out error);
    115           if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during R² calculation.");
    116           originalR2 = originalR2 * originalR2;
    117           break;
    118         case DataPartitionEnum.Training:
    119           rows = problemData.TrainingIndices;
    120           targetValues = problemData.TargetVariableTrainingValues.ToList();
    121           originalR2 = solution.TrainingRSquared;
    122           break;
    123         case DataPartitionEnum.Test:
    124           rows = problemData.TestIndices;
    125           targetValues = problemData.TargetVariableTestValues.ToList();
    126           originalR2 = solution.TestRSquared;
    127           break;
    128         default: throw new ArgumentException(string.Format("DataPartition {0} cannot be handled.", data));
    129       }
     111      DataPartitionEnum dataPartition = DataPartitionEnum.Training) {
     112
     113      IEnumerable<int> rows = GetPartitionRows(dataPartition, solution.ProblemData);
     114      IEnumerable<double> estimatedValues = solution.GetEstimatedValues(rows);
     115      return CalculateImpacts(solution.Model, solution.ProblemData, estimatedValues, rows, replacementMethod, factorReplacementMethod);
     116    }
     117
     118    public static IEnumerable<Tuple<string, double>> CalculateImpacts(
     119     IRegressionModel model,
     120     IRegressionProblemData problemData,
     121     IEnumerable<double> estimatedValues,
     122     IEnumerable<int> rows,
     123     ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
     124     FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
     125
     126      //fholzing: try and catch in case a different dataset is loaded, otherwise statement is neglectable
     127      var missingVariables = model.VariablesUsedForPrediction.Except(problemData.Dataset.VariableNames);
     128      if (missingVariables.Any()) {
     129        throw new InvalidOperationException(string.Format("Can not calculate variable impacts, because the model uses inputs missing in the dataset ({0})", string.Join(", ", missingVariables)));
     130      }
     131      IEnumerable<double> targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
     132      var originalQuality = CalculateQuality(targetValues, estimatedValues);
    130133
    131134      var impacts = new Dictionary<string, double>();
    132       var modifiableDataset = ((Dataset)dataset).ToModifiable();
    133 
    134       var inputvariables = new HashSet<string>(problemData.AllowedInputVariables.Union(solution.Model.VariablesUsedForPrediction));
    135       var allowedInputVariables = dataset.VariableNames.Where(v => inputvariables.Contains(v)).ToList();
    136 
    137       int curIdx = 0;
    138       int count = allowedInputVariables.Where(problemData.Dataset.VariableHasType<double>).Count();
    139       // calculate impacts for double variables
    140       foreach (var inputVariable in allowedInputVariables.Where(problemData.Dataset.VariableHasType<double>)) {
    141         //Report the current progress in percent. If the callback returns true, it means the execution shall be stopped
    142         if (progressCallback != null) {
    143           curIdx++;
    144           if (progressCallback((double)curIdx / count, string.Format("Calculating impact for variable {0} ({1} of {2})", inputVariable, curIdx, count))) { return null; }
    145         }
    146         var newEstimates = EvaluateModelWithReplacedVariable(solution.Model, inputVariable, modifiableDataset, rows, replacementMethod);
    147         var newR2 = OnlinePearsonsRCalculator.Calculate(targetValues, newEstimates, out error);
    148         if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during R² calculation with replaced inputs.");
    149 
    150         newR2 = newR2 * newR2;
    151         var impact = originalR2 - newR2;
    152         impacts[inputVariable] = impact;
    153       }
    154 
    155       // calculate impacts for string variables
    156       foreach (var inputVariable in allowedInputVariables.Where(problemData.Dataset.VariableHasType<string>)) {
    157         if (factorReplacementMethod == FactorReplacementMethodEnum.Best) {
    158           // try replacing with all possible values and find the best replacement value
    159           var smallestImpact = double.PositiveInfinity;
    160           foreach (var repl in problemData.Dataset.GetStringValues(inputVariable, rows).Distinct()) {
    161             var newEstimates = EvaluateModelWithReplacedVariable(solution.Model, inputVariable, modifiableDataset, rows,
    162               Enumerable.Repeat(repl, dataset.Rows));
    163             var newR2 = OnlinePearsonsRCalculator.Calculate(targetValues, newEstimates, out error);
    164             if (error != OnlineCalculatorError.None)
    165               throw new InvalidOperationException("Error during R² calculation with replaced inputs.");
    166 
    167             newR2 = newR2 * newR2;
    168             var impact = originalR2 - newR2;
    169             if (impact < smallestImpact) smallestImpact = impact;
    170           }
    171           impacts[inputVariable] = smallestImpact;
    172         } else {
    173           // for replacement methods shuffle and mode
    174           // calculate impacts for factor variables
    175 
    176           var newEstimates = EvaluateModelWithReplacedVariable(solution.Model, inputVariable, modifiableDataset, rows,
    177             factorReplacementMethod);
    178           var newR2 = OnlinePearsonsRCalculator.Calculate(targetValues, newEstimates, out error);
    179           if (error != OnlineCalculatorError.None)
    180             throw new InvalidOperationException("Error during R² calculation with replaced inputs.");
    181 
    182           newR2 = newR2 * newR2;
    183           var impact = originalR2 - newR2;
    184           impacts[inputVariable] = impact;
    185         }
    186       } // foreach
    187       return impacts.OrderByDescending(i => i.Value).Select(i => Tuple.Create(i.Key, i.Value));
    188     }
    189 
    190 
    191     private static IEnumerable<double> EvaluateModelWithReplacedVariable(IRegressionModel model, string variable, ModifiableDataset dataset, IEnumerable<int> rows, ReplacementMethodEnum replacement = ReplacementMethodEnum.Median) {
    192       var originalValues = dataset.GetReadOnlyDoubleValues(variable).ToList();
     135      var inputvariables = new HashSet<string>(problemData.AllowedInputVariables.Union(model.VariablesUsedForPrediction));
     136      var modifiableDataset = ((Dataset)(problemData.Dataset).Clone()).ToModifiable();
     137
     138      foreach (var inputVariable in inputvariables) {
     139        impacts[inputVariable] = CalculateImpact(inputVariable, model, problemData, modifiableDataset, rows, replacementMethod, factorReplacementMethod, targetValues, originalQuality);
     140      }
     141
     142      return impacts.Select(i => Tuple.Create(i.Key, i.Value));
     143    }
     144
     145    public static double CalculateImpact(string variableName,
     146      IRegressionModel model,
     147      IRegressionProblemData problemData,
     148      ModifiableDataset modifiableDataset,
     149      IEnumerable<int> rows,
     150      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
     151      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
     152      IEnumerable<double> targetValues = null,
     153      double quality = double.NaN) {
     154
     155      if (!model.VariablesUsedForPrediction.Contains(variableName)) { return 0.0; }
     156      if (!problemData.Dataset.VariableNames.Contains(variableName)) {
     157        throw new InvalidOperationException(string.Format("Can not calculate variable impact, because the model uses inputs missing in the dataset ({0})", variableName));
     158      }
     159
     160      if (targetValues == null) {
     161        targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
     162      }
     163      if (quality == double.NaN) {
     164        quality = CalculateQuality(model.GetEstimatedValues(modifiableDataset, rows), targetValues);
     165      }
     166
     167      IList originalValues = null;
     168      IList replacementValues = GetReplacementValues(modifiableDataset, variableName, model, rows, targetValues, out originalValues, replacementMethod, factorReplacementMethod);
     169
     170      double newValue = CalculateQualityForReplacement(model, modifiableDataset, variableName, originalValues, rows, replacementValues, targetValues);
     171      double impact = quality - newValue;
     172
     173      return impact;
     174    }
     175
     176    private static IList GetReplacementValues(ModifiableDataset modifiableDataset,
     177      string variableName,
     178      IRegressionModel model,
     179      IEnumerable<int> rows,
     180      IEnumerable<double> targetValues,
     181      out IList originalValues,
     182      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
     183      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
     184
     185      IList replacementValues = null;
     186      if (modifiableDataset.VariableHasType<double>(variableName)) {
     187        originalValues = modifiableDataset.GetReadOnlyDoubleValues(variableName).ToList();
     188        replacementValues = GetReplacementValuesForDouble(modifiableDataset, rows, (List<double>)originalValues, replacementMethod);
     189      } else if (modifiableDataset.VariableHasType<string>(variableName)) {
     190        originalValues = modifiableDataset.GetReadOnlyStringValues(variableName).ToList();
     191        replacementValues = GetReplacementValuesForString(model, modifiableDataset, variableName, rows, (List<string>)originalValues, targetValues, factorReplacementMethod);
     192      } else {
     193        throw new NotSupportedException("Variable not supported");
     194      }
     195
     196      return replacementValues;
     197    }
     198
     199    private static IList GetReplacementValuesForDouble(ModifiableDataset modifiableDataset,
     200      IEnumerable<int> rows,
     201      List<double> originalValues,
     202      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle) {
     203
     204      IRandom random = new FastRandom(31415);
     205      List<double> replacementValues;
    193206      double replacementValue;
    194       List<double> replacementValues;
    195       IRandom rand;
    196 
    197       switch (replacement) {
     207
     208      switch (replacementMethod) {
    198209        case ReplacementMethodEnum.Median:
    199210          replacementValue = rows.Select(r => originalValues[r]).Median();
    200           replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
     211          replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
    201212          break;
    202213        case ReplacementMethodEnum.Average:
    203214          replacementValue = rows.Select(r => originalValues[r]).Average();
    204           replacementValues = Enumerable.Repeat(replacementValue, dataset.Rows).ToList();
     215          replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
    205216          break;
    206217        case ReplacementMethodEnum.Shuffle:
    207218          // new var has same empirical distribution but the relation to y is broken
    208           rand = new FastRandom(31415);
    209219          // prepare a complete column for the dataset
    210           replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
     220          replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
    211221          // shuffle only the selected rows
    212           var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
     222          var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
    213223          int i = 0;
    214224          // update column values
     
    220230          var avg = rows.Select(r => originalValues[r]).Average();
    221231          var stdDev = rows.Select(r => originalValues[r]).StandardDeviation();
    222           rand = new FastRandom(31415);
    223232          // prepare a complete column for the dataset
    224           replacementValues = Enumerable.Repeat(double.NaN, dataset.Rows).ToList();
     233          replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
    225234          // update column values
    226235          foreach (var r in rows) {
    227             replacementValues[r] = NormalDistributedRandom.NextDouble(rand, avg, stdDev);
     236            replacementValues[r] = NormalDistributedRandom.NextDouble(random, avg, stdDev);
    228237          }
    229238          break;
    230239
    231240        default:
    232           throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacement));
    233       }
    234 
    235       return EvaluateModelWithReplacedVariable(model, variable, dataset, rows, replacementValues);
    236     }
    237 
    238     private static IEnumerable<double> EvaluateModelWithReplacedVariable(
    239       IRegressionModel model, string variable, ModifiableDataset dataset,
    240       IEnumerable<int> rows,
    241       FactorReplacementMethodEnum replacement = FactorReplacementMethodEnum.Shuffle) {
    242       var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
    243       List<string> replacementValues;
    244       IRandom rand;
    245 
    246       switch (replacement) {
     241          throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacementMethod));
     242      }
     243
     244      return replacementValues;
     245    }
     246
     247    private static IList GetReplacementValuesForString(IRegressionModel model,
     248      ModifiableDataset modifiableDataset,
     249      string variableName,
     250      IEnumerable<int> rows,
     251      List<string> originalValues,
     252      IEnumerable<double> targetValues,
     253      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Shuffle) {
     254
     255      List<string> replacementValues = null;
     256      IRandom random = new FastRandom(31415);
     257
     258      switch (factorReplacementMethod) {
     259        case FactorReplacementMethodEnum.Best:
     260          // try replacing with all possible values and find the best replacement value
     261          var bestQuality = double.NegativeInfinity;
     262          foreach (var repl in modifiableDataset.GetStringValues(variableName, rows).Distinct()) {
     263            List<string> curReplacementValues = Enumerable.Repeat(repl, modifiableDataset.Rows).ToList();
     264            //fholzing: this result could be used later on (theoretically), but is neglected for better readability/method consistency
     265            var newValue = CalculateQualityForReplacement(model, modifiableDataset, variableName, originalValues, rows, curReplacementValues, targetValues);
     266            var curQuality = newValue;
     267
     268            if (curQuality > bestQuality) {
     269              bestQuality = curQuality;
     270              replacementValues = curReplacementValues;
     271            }
     272          }
     273          break;
    247274        case FactorReplacementMethodEnum.Mode:
    248275          var mostCommonValue = rows.Select(r => originalValues[r])
     
    250277            .OrderByDescending(g => g.Count())
    251278            .First().Key;
    252           replacementValues = Enumerable.Repeat(mostCommonValue, dataset.Rows).ToList();
     279          replacementValues = Enumerable.Repeat(mostCommonValue, modifiableDataset.Rows).ToList();
    253280          break;
    254281        case FactorReplacementMethodEnum.Shuffle:
    255282          // new var has same empirical distribution but the relation to y is broken
    256           rand = new FastRandom(31415);
    257283          // prepare a complete column for the dataset
    258           replacementValues = Enumerable.Repeat(string.Empty, dataset.Rows).ToList();
     284          replacementValues = Enumerable.Repeat(string.Empty, modifiableDataset.Rows).ToList();
    259285          // shuffle only the selected rows
    260           var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
     286          var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
    261287          int i = 0;
    262288          // update column values
     
    266292          break;
    267293        default:
    268           throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", replacement));
    269       }
    270 
    271       return EvaluateModelWithReplacedVariable(model, variable, dataset, rows, replacementValues);
    272     }
    273 
    274     private static IEnumerable<double> EvaluateModelWithReplacedVariable(IRegressionModel model, string variable,
    275       ModifiableDataset dataset, IEnumerable<int> rows, IEnumerable<double> replacementValues) {
    276       var originalValues = dataset.GetReadOnlyDoubleValues(variable).ToList();
    277       dataset.ReplaceVariable(variable, replacementValues.ToList());
     294          throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", factorReplacementMethod));
     295      }
     296
     297      return replacementValues;
     298    }
     299
     300    private static double CalculateQualityForReplacement(
     301      IRegressionModel model,
     302      ModifiableDataset modifiableDataset,
     303      string variableName,
     304      IList originalValues,
     305      IEnumerable<int> rows,
     306      IList replacementValues,
     307      IEnumerable<double> targetValues) {
     308
     309      modifiableDataset.ReplaceVariable(variableName, replacementValues);
    278310      //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
    279       var estimates = model.GetEstimatedValues(dataset, rows).ToList();
    280       dataset.ReplaceVariable(variable, originalValues);
    281 
    282       return estimates;
    283     }
    284     private static IEnumerable<double> EvaluateModelWithReplacedVariable(IRegressionModel model, string variable,
    285       ModifiableDataset dataset, IEnumerable<int> rows, IEnumerable<string> replacementValues) {
    286       var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
    287       dataset.ReplaceVariable(variable, replacementValues.ToList());
    288       //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
    289       var estimates = model.GetEstimatedValues(dataset, rows).ToList();
    290       dataset.ReplaceVariable(variable, originalValues);
    291 
    292       return estimates;
     311      var estimates = model.GetEstimatedValues(modifiableDataset, rows).ToList();
     312      var ret = CalculateQuality(targetValues, estimates);
     313      modifiableDataset.ReplaceVariable(variableName, originalValues);
     314
     315      return ret;
     316    }
     317
     318    public static double CalculateQuality(IEnumerable<double> targetValues, IEnumerable<double> estimatedValues) {
     319      OnlineCalculatorError errorState;
     320      var ret = OnlinePearsonsRCalculator.Calculate(targetValues, estimatedValues, out errorState);
     321      if (errorState != OnlineCalculatorError.None) { throw new InvalidOperationException("Error during calculation with replaced inputs."); }
     322      return ret * ret;
     323    }
     324
     325    public static IEnumerable<int> GetPartitionRows(DataPartitionEnum dataPartition, IRegressionProblemData problemData) {
     326      IEnumerable<int> rows;
     327
     328      switch (dataPartition) {
     329        case DataPartitionEnum.All:
     330          rows = problemData.AllIndices;
     331          break;
     332        case DataPartitionEnum.Test:
     333          rows = problemData.TestIndices;
     334          break;
     335        case DataPartitionEnum.Training:
     336          rows = problemData.TrainingIndices;
     337          break;
     338        default:
     339          throw new NotSupportedException("DataPartition not supported");
     340      }
     341
     342      return rows;
    293343    }
    294344  }
  • branches/2971_named_intervals/NamedIntervals.sln

    r16426 r16536  
    1919    Test.cmd = Test.cmd
    2020  EndProjectSection
     21EndProject
     22Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "HeuristicLab.Problems.DataAnalysis.Views-3.4", "HeuristicLab.Problems.DataAnalysis.Views\3.4\HeuristicLab.Problems.DataAnalysis.Views-3.4.csproj", "{3E9E8944-44FF-40BB-A622-3A4A7DD0F198}"
    2123EndProject
    2224Global
     
    5456    {72104A0B-90E7-42F3-9ABE-9BBBADD4B943}.Release|x86.ActiveCfg = Release|x86
    5557    {72104A0B-90E7-42F3-9ABE-9BBBADD4B943}.Release|x86.Build.0 = Release|x86
     58    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
     59    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Debug|Any CPU.Build.0 = Debug|Any CPU
     60    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Debug|x64.ActiveCfg = Debug|x64
     61    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Debug|x64.Build.0 = Debug|x64
     62    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Debug|x86.ActiveCfg = Debug|x86
     63    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Debug|x86.Build.0 = Debug|x86
     64    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Release|Any CPU.ActiveCfg = Release|Any CPU
     65    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Release|Any CPU.Build.0 = Release|Any CPU
     66    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Release|x64.ActiveCfg = Release|x64
     67    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Release|x64.Build.0 = Release|x64
     68    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Release|x86.ActiveCfg = Release|x86
     69    {3E9E8944-44FF-40BB-A622-3A4A7DD0F198}.Release|x86.Build.0 = Release|x86
    5670  EndGlobalSection
    5771  GlobalSection(SolutionProperties) = preSolution
Note: See TracChangeset for help on using the changeset viewer.