[15638] | 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 |
|
---|
| 24 | using System;
|
---|
[16189] | 25 | using System.Collections;
|
---|
[15638] | 26 | using System.Collections.Generic;
|
---|
| 27 | using System.Linq;
|
---|
| 28 | using HeuristicLab.Common;
|
---|
| 29 | using HeuristicLab.Core;
|
---|
| 30 | using HeuristicLab.Data;
|
---|
| 31 | using HeuristicLab.Parameters;
|
---|
| 32 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
| 33 | using HeuristicLab.Random;
|
---|
| 34 |
|
---|
| 35 | namespace 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 {
|
---|
[16189] | 39 | #region Parameters/Properties
|
---|
[15638] | 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";
|
---|
[16189] | 58 | private const string FactorReplacementParameterName = "Factor Replacement Method";
|
---|
[15638] | 59 | private const string DataPartitionParameterName = "DataPartition";
|
---|
| 60 |
|
---|
| 61 | public IFixedValueParameter<EnumValue<ReplacementMethodEnum>> ReplacementParameter {
|
---|
| 62 | get { return (IFixedValueParameter<EnumValue<ReplacementMethodEnum>>)Parameters[ReplacementParameterName]; }
|
---|
| 63 | }
|
---|
[16189] | 64 | public IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>> FactorReplacementParameter {
|
---|
| 65 | get { return (IFixedValueParameter<EnumValue<FactorReplacementMethodEnum>>)Parameters[FactorReplacementParameterName]; }
|
---|
| 66 | }
|
---|
[15638] | 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 | }
|
---|
[16189] | 75 | public FactorReplacementMethodEnum FactorReplacementMethod {
|
---|
| 76 | get { return FactorReplacementParameter.Value.Value; }
|
---|
| 77 | set { FactorReplacementParameter.Value.Value = value; }
|
---|
| 78 | }
|
---|
[15638] | 79 | public DataPartitionEnum DataPartition {
|
---|
| 80 | get { return DataPartitionParameter.Value.Value; }
|
---|
| 81 | set { DataPartitionParameter.Value.Value = value; }
|
---|
| 82 | }
|
---|
[16189] | 83 | #endregion
|
---|
[15638] | 84 |
|
---|
[16189] | 85 | #region Ctor/Cloner
|
---|
[15638] | 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() {
|
---|
[16189] | 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)));
|
---|
[15638] | 94 | Parameters.Add(new FixedValueParameter<EnumValue<DataPartitionEnum>>(DataPartitionParameterName, "The data partition on which the impacts are calculated.", new EnumValue<DataPartitionEnum>(DataPartitionEnum.Training)));
|
---|
| 95 | }
|
---|
| 96 |
|
---|
[16189] | 97 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
| 98 | return new ClassificationSolutionVariableImpactsCalculator(this, cloner);
|
---|
| 99 | }
|
---|
| 100 | #endregion
|
---|
| 101 |
|
---|
[15638] | 102 | //mkommend: annoying name clash with static method, open to better naming suggestions
|
---|
| 103 | public IEnumerable<Tuple<string, double>> Calculate(IClassificationSolution solution) {
|
---|
[16189] | 104 | return CalculateImpacts(solution, ReplacementMethod, FactorReplacementMethod, DataPartition);
|
---|
[15638] | 105 | }
|
---|
| 106 |
|
---|
| 107 | public static IEnumerable<Tuple<string, double>> CalculateImpacts(
|
---|
| 108 | IClassificationSolution solution,
|
---|
[16189] | 109 | ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Shuffle,
|
---|
| 110 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best,
|
---|
| 111 | DataPartitionEnum dataPartition = DataPartitionEnum.Training) {
|
---|
[15638] | 112 |
|
---|
[16189] | 113 | IEnumerable<int> rows = GetPartitionRows(dataPartition, solution.ProblemData);
|
---|
| 114 | IEnumerable<double> estimatedClassValues = solution.GetEstimatedClassValues(rows);
|
---|
[16190] | 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);
|
---|
[16189] | 118 | }
|
---|
[15638] | 119 |
|
---|
[16189] | 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) {
|
---|
[15638] | 127 |
|
---|
[16189] | 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)));
|
---|
[16055] | 132 | }
|
---|
[16189] | 133 | IEnumerable<double> targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
|
---|
| 134 | var originalQuality = CalculateQuality(targetValues, estimatedClassValues);
|
---|
[16055] | 135 |
|
---|
[15638] | 136 | var impacts = new Dictionary<string, double>();
|
---|
[16189] | 137 | var inputvariables = new HashSet<string>(problemData.AllowedInputVariables.Union(model.VariablesUsedForPrediction));
|
---|
| 138 | var modifiableDataset = ((Dataset)(problemData.Dataset).Clone()).ToModifiable();
|
---|
[15638] | 139 |
|
---|
[16189] | 140 | foreach (var inputVariable in inputvariables) {
|
---|
| 141 | impacts[inputVariable] = CalculateImpact(inputVariable, model, problemData, modifiableDataset, rows, replacementMethod, factorReplacementMethod, targetValues, originalQuality);
|
---|
| 142 | }
|
---|
[15638] | 143 |
|
---|
[16189] | 144 | return impacts.Select(i => Tuple.Create(i.Key, i.Value));
|
---|
| 145 | }
|
---|
[15638] | 146 |
|
---|
[16189] | 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));
|
---|
[16055] | 160 | }
|
---|
| 161 |
|
---|
[16189] | 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 | }
|
---|
[16055] | 168 |
|
---|
[16189] | 169 | IList originalValues = null;
|
---|
| 170 | IList replacementValues = GetReplacementValues(modifiableDataset, variableName, model, rows, targetValues, out originalValues, replacementMethod, factorReplacementMethod);
|
---|
[16055] | 171 |
|
---|
[16189] | 172 | double newValue = CalculateQualityForReplacement(model, modifiableDataset, variableName, originalValues, rows, replacementValues, targetValues);
|
---|
| 173 | double impact = quality - newValue;
|
---|
[16055] | 174 |
|
---|
[16189] | 175 | return impact;
|
---|
[16055] | 176 | }
|
---|
| 177 |
|
---|
[16189] | 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,
|
---|
| 185 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
|
---|
| 186 |
|
---|
| 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;
|
---|
[16188] | 208 | double replacementValue;
|
---|
[15638] | 209 |
|
---|
[16189] | 210 | switch (replacementMethod) {
|
---|
[16055] | 211 | case ReplacementMethodEnum.Median:
|
---|
| 212 | replacementValue = rows.Select(r => originalValues[r]).Median();
|
---|
[16189] | 213 | replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
|
---|
[16055] | 214 | break;
|
---|
| 215 | case ReplacementMethodEnum.Average:
|
---|
| 216 | replacementValue = rows.Select(r => originalValues[r]).Average();
|
---|
[16189] | 217 | replacementValues = Enumerable.Repeat(replacementValue, modifiableDataset.Rows).ToList();
|
---|
[16055] | 218 | break;
|
---|
| 219 | case ReplacementMethodEnum.Shuffle:
|
---|
| 220 | // new var has same empirical distribution but the relation to y is broken
|
---|
| 221 | // prepare a complete column for the dataset
|
---|
[16189] | 222 | replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
|
---|
[16055] | 223 | // shuffle only the selected rows
|
---|
[16189] | 224 | var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
|
---|
[16055] | 225 | int i = 0;
|
---|
| 226 | // update column values
|
---|
| 227 | foreach (var r in rows) {
|
---|
| 228 | replacementValues[r] = shuffledValues[i++];
|
---|
| 229 | }
|
---|
| 230 | break;
|
---|
| 231 | case ReplacementMethodEnum.Noise:
|
---|
| 232 | var avg = rows.Select(r => originalValues[r]).Average();
|
---|
| 233 | var stdDev = rows.Select(r => originalValues[r]).StandardDeviation();
|
---|
| 234 | // prepare a complete column for the dataset
|
---|
[16189] | 235 | replacementValues = Enumerable.Repeat(double.NaN, modifiableDataset.Rows).ToList();
|
---|
[16055] | 236 | // update column values
|
---|
| 237 | foreach (var r in rows) {
|
---|
[16189] | 238 | replacementValues[r] = NormalDistributedRandom.NextDouble(random, avg, stdDev);
|
---|
[16055] | 239 | }
|
---|
| 240 | break;
|
---|
[15638] | 241 |
|
---|
[16055] | 242 | default:
|
---|
[16189] | 243 | throw new ArgumentException(string.Format("ReplacementMethod {0} cannot be handled.", replacementMethod));
|
---|
[16055] | 244 | }
|
---|
[16036] | 245 |
|
---|
[16189] | 246 | return replacementValues;
|
---|
[16055] | 247 | }
|
---|
[16036] | 248 |
|
---|
[16189] | 249 | private static IList GetReplacementValuesForString(IClassificationModel model,
|
---|
| 250 | ModifiableDataset modifiableDataset,
|
---|
| 251 | string variableName,
|
---|
[16055] | 252 | IEnumerable<int> rows,
|
---|
[16189] | 253 | List<string> originalValues,
|
---|
| 254 | IEnumerable<double> targetValues,
|
---|
| 255 | FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Shuffle) {
|
---|
[16036] | 256 |
|
---|
[16189] | 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;
|
---|
[16055] | 276 | case FactorReplacementMethodEnum.Mode:
|
---|
| 277 | var mostCommonValue = rows.Select(r => originalValues[r])
|
---|
| 278 | .GroupBy(v => v)
|
---|
| 279 | .OrderByDescending(g => g.Count())
|
---|
| 280 | .First().Key;
|
---|
[16189] | 281 | replacementValues = Enumerable.Repeat(mostCommonValue, modifiableDataset.Rows).ToList();
|
---|
[16055] | 282 | break;
|
---|
| 283 | case FactorReplacementMethodEnum.Shuffle:
|
---|
| 284 | // new var has same empirical distribution but the relation to y is broken
|
---|
| 285 | // prepare a complete column for the dataset
|
---|
[16189] | 286 | replacementValues = Enumerable.Repeat(string.Empty, modifiableDataset.Rows).ToList();
|
---|
[16055] | 287 | // shuffle only the selected rows
|
---|
[16189] | 288 | var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(random).ToList();
|
---|
[16055] | 289 | int i = 0;
|
---|
| 290 | // update column values
|
---|
| 291 | foreach (var r in rows) {
|
---|
| 292 | replacementValues[r] = shuffledValues[i++];
|
---|
| 293 | }
|
---|
| 294 | break;
|
---|
| 295 | default:
|
---|
[16189] | 296 | throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", factorReplacementMethod));
|
---|
[15638] | 297 | }
|
---|
| 298 |
|
---|
[16189] | 299 | return replacementValues;
|
---|
[15638] | 300 | }
|
---|
| 301 |
|
---|
[16189] | 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);
|
---|
| 312 | var discModel = model as IDiscriminantFunctionClassificationModel;
|
---|
| 313 | if (discModel != null) {
|
---|
| 314 | var problemData = new ClassificationProblemData(modifiableDataset, modifiableDataset.VariableNames, model.TargetVariable);
|
---|
| 315 | discModel.RecalculateModelParameters(problemData, rows);
|
---|
| 316 | }
|
---|
| 317 |
|
---|
[16037] | 318 | //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
|
---|
[16189] | 319 | var estimates = model.GetEstimatedClassValues(modifiableDataset, rows).ToList();
|
---|
| 320 | var ret = CalculateQuality(targetValues, estimates);
|
---|
| 321 | modifiableDataset.ReplaceVariable(variableName, originalValues);
|
---|
[16037] | 322 |
|
---|
[16189] | 323 | return ret;
|
---|
[16037] | 324 | }
|
---|
| 325 |
|
---|
[16189] | 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;
|
---|
[15638] | 331 | }
|
---|
[16189] | 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;
|
---|
| 351 | }
|
---|
[15638] | 352 | }
|
---|
[16189] | 353 | } |
---|