Free cookie consent management tool by TermsFeed Policy Generator

source: branches/2904_CalculateImpacts/3.4/Implementation/Classification/ClassificationSolutionVariableImpactsCalculator.cs @ 16189

Last change on this file since 16189 was 16189, checked in by fholzing, 6 years ago

#2904: Overwrote trunk-changes with branch-changes

File size: 16.6 KB
Line 
1#region License Information
2
3/* HeuristicLab
4 * Copyright (C) 2002-2018 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
5 *
6 * This file is part of HeuristicLab.
7 *
8 * HeuristicLab is free software: you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation, either version 3 of the License, or
11 * (at your option) any later version.
12 *
13 * HeuristicLab is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
20 */
21
22#endregion
23
24using System;
25using System.Collections;
26using System.Collections.Generic;
27using System.Linq;
28using HeuristicLab.Common;
29using HeuristicLab.Core;
30using HeuristicLab.Data;
31using HeuristicLab.Parameters;
32using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
33using HeuristicLab.Random;
34
35namespace HeuristicLab.Problems.DataAnalysis {
36  [StorableClass]
37  [Item("ClassificationSolution Impacts Calculator", "Calculation of the impacts of input variables for any classification solution")]
38  public sealed class ClassificationSolutionVariableImpactsCalculator : ParameterizedNamedItem {
39    #region Parameters/Properties
40    public enum ReplacementMethodEnum {
41      Median,
42      Average,
43      Shuffle,
44      Noise
45    }
46    public enum FactorReplacementMethodEnum {
47      Best,
48      Mode,
49      Shuffle
50    }
51    public enum DataPartitionEnum {
52      Training,
53      Test,
54      All
55    }
56
57    private const string ReplacementParameterName = "Replacement Method";
58    private const string FactorReplacementParameterName = "Factor Replacement Method";
59    private const string DataPartitionParameterName = "DataPartition";
60
61    public IFixedValueParameter<EnumValue<ReplacementMethodEnum>> ReplacementParameter {
62      get { return (IFixedValueParameter<EnumValue<ReplacementMethodEnum>>)Parameters[ReplacementParameterName]; }
63    }
64    public IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>> FactorReplacementParameter {
65      get { return (IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>>)Parameters[FactorReplacementParameterName]; }
66    }
67    public IFixedValueParameter<EnumValue<DataPartitionEnum>> DataPartitionParameter {
68      get { return (IFixedValueParameter<EnumValue<DataPartitionEnum>>)Parameters[DataPartitionParameterName]; }
69    }
70
71    public ReplacementMethodEnum ReplacementMethod {
72      get { return ReplacementParameter.Value.Value; }
73      set { ReplacementParameter.Value.Value = value; }
74    }
75    public FactorReplacementMethodEnum FactorReplacementMethod {
76      get { return FactorReplacementParameter.Value.Value; }
77      set { FactorReplacementParameter.Value.Value = value; }
78    }
79    public DataPartitionEnum DataPartition {
80      get { return DataPartitionParameter.Value.Value; }
81      set { DataPartitionParameter.Value.Value = value; }
82    }
83    #endregion
84
85    #region Ctor/Cloner
86    [StorableConstructor]
87    private ClassificationSolutionVariableImpactsCalculator(bool deserializing) : base(deserializing) { }
88    private ClassificationSolutionVariableImpactsCalculator(ClassificationSolutionVariableImpactsCalculator original, Cloner cloner)
89      : 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
97    public override IDeepCloneable Clone(Cloner cloner) {
98      return new ClassificationSolutionVariableImpactsCalculator(this, cloner);
99    }
100    #endregion
101
102    //mkommend: annoying name clash with static method, open to better naming suggestions
103    public IEnumerable<Tuple<string, double>> Calculate(IClassificationSolution solution) {
104      return CalculateImpacts(solution, ReplacementMethod, FactorReplacementMethod, DataPartition);
105    }
106
107    public static IEnumerable<Tuple<string, double>> CalculateImpacts(
108      IClassificationSolution solution,
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      return CalculateImpacts(solution.Model, solution.ProblemData, estimatedClassValues, rows, replacementMethod, factorReplacementMethod);
116    }
117
118    public static IEnumerable<Tuple<string, double>> CalculateImpacts(
119     IClassificationModel model,
120     IClassificationProblemData problemData,
121     IEnumerable<double> estimatedClassValues,
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, estimatedClassValues);
133
134      var impacts = new Dictionary<string, double>();
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      IClassificationModel model,
147      IClassificationProblemData 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.GetEstimatedClassValues(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      IClassificationModel 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;
206      double replacementValue;
207
208      switch (replacementMethod) {
209        case ReplacementMethodEnum.Median:
210          replacementValue = rows.Select(r => originalValues[r]).Median();
211          replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
212          break;
213        case ReplacementMethodEnum.Average:
214          replacementValue = rows.Select(r => originalValues[r]).Average();
215          replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
216          break;
217        case ReplacementMethodEnum.Shuffle:
218          // new var has same empirical distribution but the relation to y is broken
219          // prepare a complete column for the dataset
220          replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
221          // shuffle only the selected rows
222          var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
223          int i = 0;
224          // update column values
225          foreach (var r in rows) {
226            replacementValues[r] = shuffledValues[i++];
227          }
228          break;
229        case ReplacementMethodEnum.Noise:
230          var avg = rows.Select(r => originalValues[r]).Average();
231          var stdDev = rows.Select(r => originalValues[r]).StandardDeviation();
232          // prepare a complete column for the dataset
233          replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
234          // update column values
235          foreach (var r in rows) {
236            replacementValues[r] = NormalDistributedRandom.NextDouble(random, avg, stdDev);
237          }
238          break;
239
240        default:
241          throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacementMethod));
242      }
243
244      return replacementValues;
245    }
246
247    private static IList GetReplacementValuesForString(IClassificationModel 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;
274        case FactorReplacementMethodEnum.Mode:
275          var mostCommonValue = rows.Select(r => originalValues[r])
276            .GroupBy(v => v)
277            .OrderByDescending(g => g.Count())
278            .First().Key;
279          replacementValues = Enumerable.Repeat(mostCommonValue, modifiableDataset.Rows).ToList();
280          break;
281        case FactorReplacementMethodEnum.Shuffle:
282          // new var has same empirical distribution but the relation to y is broken
283          // prepare a complete column for the dataset
284          replacementValues = Enumerable.Repeat(string.Empty, modifiableDataset.Rows).ToList();
285          // shuffle only the selected rows
286          var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
287          int i = 0;
288          // update column values
289          foreach (var r in rows) {
290            replacementValues[r] = shuffledValues[i++];
291          }
292          break;
293        default:
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      IClassificationModel 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);
310      var discModel = model as IDiscriminantFunctionClassificationModel;
311      if (discModel != null) {
312        var problemData = new ClassificationProblemData(modifiableDataset, modifiableDataset.VariableNames, model.TargetVariable);
313        discModel.RecalculateModelParameters(problemData, rows);
314      }
315
316      //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
317      var estimates = model.GetEstimatedClassValues(modifiableDataset, rows).ToList();
318      var ret = CalculateQuality(targetValues, estimates);
319      modifiableDataset.ReplaceVariable(variableName, originalValues);
320
321      return ret;
322    }
323
324    public static double CalculateQuality(IEnumerable<double> targetValues, IEnumerable<double> estimatedClassValues) {
325      OnlineCalculatorError errorState;
326      var ret = OnlineAccuracyCalculator.Calculate(targetValues, estimatedClassValues, out errorState);
327      if (errorState != OnlineCalculatorError.None) { throw new InvalidOperationException("Error during calculation with replaced inputs."); }
328      return ret;
329    }
330
331    public static IEnumerable<int> GetPartitionRows(DataPartitionEnum dataPartition, IClassificationProblemData problemData) {
332      IEnumerable<int> rows;
333
334      switch (dataPartition) {
335        case DataPartitionEnum.All:
336          rows = problemData.AllIndices;
337          break;
338        case DataPartitionEnum.Test:
339          rows = problemData.TestIndices;
340          break;
341        case DataPartitionEnum.Training:
342          rows = problemData.TrainingIndices;
343          break;
344        default:
345          throw new NotSupportedException("DataPartition not supported");
346      }
347
348      return rows;
349    }
350  }
351}
Note: See TracBrowser for help on using the repository browser.