1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2011 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
4 | *
|
---|
5 | * This file is part of HeuristicLab.
|
---|
6 | *
|
---|
7 | * HeuristicLab is free software: you can redistribute it and/or modify
|
---|
8 | * it under the terms of the GNU General Public License as published by
|
---|
9 | * the Free Software Foundation, either version 3 of the License, or
|
---|
10 | * (at your option) any later version.
|
---|
11 | *
|
---|
12 | * HeuristicLab is distributed in the hope that it will be useful,
|
---|
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
---|
15 | * GNU General Public License for more details.
|
---|
16 | *
|
---|
17 | * You should have received a copy of the GNU General Public License
|
---|
18 | * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
|
---|
19 | */
|
---|
20 | #endregion
|
---|
21 |
|
---|
22 | using System;
|
---|
23 | using System.Collections.Generic;
|
---|
24 | using System.Linq;
|
---|
25 | using HeuristicLab.Collections;
|
---|
26 | using HeuristicLab.Common;
|
---|
27 | using HeuristicLab.Core;
|
---|
28 | using HeuristicLab.Data;
|
---|
29 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
30 |
|
---|
31 | namespace HeuristicLab.Problems.DataAnalysis {
|
---|
32 | /// <summary>
|
---|
33 | /// Represents classification solutions that contain an ensemble of multiple classification models
|
---|
34 | /// </summary>
|
---|
35 | [StorableClass]
|
---|
36 | [Item("Classification Ensemble Solution", "A classification solution that contains an ensemble of multiple classification models")]
|
---|
37 | [Creatable("Data Analysis - Ensembles")]
|
---|
38 | public sealed class ClassificationEnsembleSolution : ClassificationSolution, IClassificationEnsembleSolution {
|
---|
39 | public new IClassificationEnsembleModel Model {
|
---|
40 | get { return (IClassificationEnsembleModel)base.Model; }
|
---|
41 | }
|
---|
42 | public new ClassificationEnsembleProblemData ProblemData {
|
---|
43 | get { return (ClassificationEnsembleProblemData)base.ProblemData; }
|
---|
44 | set { base.ProblemData = value; }
|
---|
45 | }
|
---|
46 |
|
---|
47 | private readonly ItemCollection<IClassificationSolution> classificationSolutions;
|
---|
48 | public IItemCollection<IClassificationSolution> ClassificationSolutions {
|
---|
49 | get { return classificationSolutions; }
|
---|
50 | }
|
---|
51 |
|
---|
52 | [Storable]
|
---|
53 | private Dictionary<IClassificationModel, IntRange> trainingPartitions;
|
---|
54 | [Storable]
|
---|
55 | private Dictionary<IClassificationModel, IntRange> testPartitions;
|
---|
56 |
|
---|
57 | [StorableConstructor]
|
---|
58 | private ClassificationEnsembleSolution(bool deserializing)
|
---|
59 | : base(deserializing) {
|
---|
60 | classificationSolutions = new ItemCollection<IClassificationSolution>();
|
---|
61 | }
|
---|
62 | [StorableHook(HookType.AfterDeserialization)]
|
---|
63 | private void AfterDeserialization() {
|
---|
64 | foreach (var model in Model.Models) {
|
---|
65 | IClassificationProblemData problemData = (IClassificationProblemData)ProblemData.Clone();
|
---|
66 | problemData.TrainingPartition.Start = trainingPartitions[model].Start;
|
---|
67 | problemData.TrainingPartition.End = trainingPartitions[model].End;
|
---|
68 | problemData.TestPartition.Start = testPartitions[model].Start;
|
---|
69 | problemData.TestPartition.End = testPartitions[model].End;
|
---|
70 |
|
---|
71 | classificationSolutions.Add(model.CreateClassificationSolution(problemData));
|
---|
72 | }
|
---|
73 | RegisterClassificationSolutionsEventHandler();
|
---|
74 | }
|
---|
75 |
|
---|
76 | private ClassificationEnsembleSolution(ClassificationEnsembleSolution original, Cloner cloner)
|
---|
77 | : base(original, cloner) {
|
---|
78 | trainingPartitions = new Dictionary<IClassificationModel, IntRange>();
|
---|
79 | testPartitions = new Dictionary<IClassificationModel, IntRange>();
|
---|
80 | foreach (var pair in original.trainingPartitions) {
|
---|
81 | trainingPartitions[cloner.Clone(pair.Key)] = cloner.Clone(pair.Value);
|
---|
82 | }
|
---|
83 | foreach (var pair in original.testPartitions) {
|
---|
84 | testPartitions[cloner.Clone(pair.Key)] = cloner.Clone(pair.Value);
|
---|
85 | }
|
---|
86 |
|
---|
87 | classificationSolutions = cloner.Clone(original.classificationSolutions);
|
---|
88 | RegisterClassificationSolutionsEventHandler();
|
---|
89 | }
|
---|
90 |
|
---|
91 | public ClassificationEnsembleSolution()
|
---|
92 | : base(new ClassificationEnsembleModel(), ClassificationEnsembleProblemData.EmptyProblemData) {
|
---|
93 | trainingPartitions = new Dictionary<IClassificationModel, IntRange>();
|
---|
94 | testPartitions = new Dictionary<IClassificationModel, IntRange>();
|
---|
95 | classificationSolutions = new ItemCollection<IClassificationSolution>();
|
---|
96 |
|
---|
97 | RegisterClassificationSolutionsEventHandler();
|
---|
98 | }
|
---|
99 |
|
---|
100 | public ClassificationEnsembleSolution(IEnumerable<IClassificationModel> models, IClassificationProblemData problemData)
|
---|
101 | : this(models, problemData,
|
---|
102 | models.Select(m => (IntRange)problemData.TrainingPartition.Clone()),
|
---|
103 | models.Select(m => (IntRange)problemData.TestPartition.Clone())
|
---|
104 | ) { }
|
---|
105 |
|
---|
106 | public ClassificationEnsembleSolution(IEnumerable<IClassificationModel> models, IClassificationProblemData problemData, IEnumerable<IntRange> trainingPartitions, IEnumerable<IntRange> testPartitions)
|
---|
107 | : base(new ClassificationEnsembleModel(Enumerable.Empty<IClassificationModel>()), new ClassificationEnsembleProblemData(problemData)) {
|
---|
108 | this.trainingPartitions = new Dictionary<IClassificationModel, IntRange>();
|
---|
109 | this.testPartitions = new Dictionary<IClassificationModel, IntRange>();
|
---|
110 | this.classificationSolutions = new ItemCollection<IClassificationSolution>();
|
---|
111 |
|
---|
112 | List<IClassificationSolution> solutions = new List<IClassificationSolution>();
|
---|
113 | var modelEnumerator = models.GetEnumerator();
|
---|
114 | var trainingPartitionEnumerator = trainingPartitions.GetEnumerator();
|
---|
115 | var testPartitionEnumerator = testPartitions.GetEnumerator();
|
---|
116 |
|
---|
117 | while (modelEnumerator.MoveNext() & trainingPartitionEnumerator.MoveNext() & testPartitionEnumerator.MoveNext()) {
|
---|
118 | var p = (IClassificationProblemData)problemData.Clone();
|
---|
119 | p.TrainingPartition.Start = trainingPartitionEnumerator.Current.Start;
|
---|
120 | p.TrainingPartition.End = trainingPartitionEnumerator.Current.End;
|
---|
121 | p.TestPartition.Start = testPartitionEnumerator.Current.Start;
|
---|
122 | p.TestPartition.End = testPartitionEnumerator.Current.End;
|
---|
123 |
|
---|
124 | solutions.Add(modelEnumerator.Current.CreateClassificationSolution(p));
|
---|
125 | }
|
---|
126 | if (modelEnumerator.MoveNext() | trainingPartitionEnumerator.MoveNext() | testPartitionEnumerator.MoveNext()) {
|
---|
127 | throw new ArgumentException();
|
---|
128 | }
|
---|
129 |
|
---|
130 | RegisterClassificationSolutionsEventHandler();
|
---|
131 | classificationSolutions.AddRange(solutions);
|
---|
132 | }
|
---|
133 |
|
---|
134 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
135 | return new ClassificationEnsembleSolution(this, cloner);
|
---|
136 | }
|
---|
137 | private void RegisterClassificationSolutionsEventHandler() {
|
---|
138 | classificationSolutions.ItemsAdded += new CollectionItemsChangedEventHandler<IClassificationSolution>(classificationSolutions_ItemsAdded);
|
---|
139 | classificationSolutions.ItemsRemoved += new CollectionItemsChangedEventHandler<IClassificationSolution>(classificationSolutions_ItemsRemoved);
|
---|
140 | classificationSolutions.CollectionReset += new CollectionItemsChangedEventHandler<IClassificationSolution>(classificationSolutions_CollectionReset);
|
---|
141 | }
|
---|
142 |
|
---|
143 | protected override void RecalculateResults() {
|
---|
144 | CalculateResults();
|
---|
145 | }
|
---|
146 |
|
---|
147 | #region Evaluation
|
---|
148 | public override IEnumerable<double> EstimatedTrainingClassValues {
|
---|
149 | get {
|
---|
150 | var rows = ProblemData.TrainingIndizes;
|
---|
151 | var estimatedValuesEnumerators = (from model in Model.Models
|
---|
152 | select new { Model = model, EstimatedValuesEnumerator = model.GetEstimatedClassValues(ProblemData.Dataset, rows).GetEnumerator() })
|
---|
153 | .ToList();
|
---|
154 | var rowsEnumerator = rows.GetEnumerator();
|
---|
155 | // aggregate to make sure that MoveNext is called for all enumerators
|
---|
156 | while (rowsEnumerator.MoveNext() & estimatedValuesEnumerators.Select(en => en.EstimatedValuesEnumerator.MoveNext()).Aggregate(true, (acc, b) => acc & b)) {
|
---|
157 | int currentRow = rowsEnumerator.Current;
|
---|
158 |
|
---|
159 | var selectedEnumerators = from pair in estimatedValuesEnumerators
|
---|
160 | where RowIsTrainingForModel(currentRow, pair.Model) && !RowIsTestForModel(currentRow, pair.Model)
|
---|
161 | select pair.EstimatedValuesEnumerator;
|
---|
162 | yield return AggregateEstimatedClassValues(selectedEnumerators.Select(x => x.Current));
|
---|
163 | }
|
---|
164 | }
|
---|
165 | }
|
---|
166 |
|
---|
167 | public override IEnumerable<double> EstimatedTestClassValues {
|
---|
168 | get {
|
---|
169 | var rows = ProblemData.TestIndizes;
|
---|
170 | var estimatedValuesEnumerators = (from model in Model.Models
|
---|
171 | select new { Model = model, EstimatedValuesEnumerator = model.GetEstimatedClassValues(ProblemData.Dataset, rows).GetEnumerator() })
|
---|
172 | .ToList();
|
---|
173 | var rowsEnumerator = ProblemData.TestIndizes.GetEnumerator();
|
---|
174 | // aggregate to make sure that MoveNext is called for all enumerators
|
---|
175 | while (rowsEnumerator.MoveNext() & estimatedValuesEnumerators.Select(en => en.EstimatedValuesEnumerator.MoveNext()).Aggregate(true, (acc, b) => acc & b)) {
|
---|
176 | int currentRow = rowsEnumerator.Current;
|
---|
177 |
|
---|
178 | var selectedEnumerators = from pair in estimatedValuesEnumerators
|
---|
179 | where RowIsTestForModel(currentRow, pair.Model)
|
---|
180 | select pair.EstimatedValuesEnumerator;
|
---|
181 |
|
---|
182 | yield return AggregateEstimatedClassValues(selectedEnumerators.Select(x => x.Current));
|
---|
183 | }
|
---|
184 | }
|
---|
185 | }
|
---|
186 |
|
---|
187 | private bool RowIsTrainingForModel(int currentRow, IClassificationModel model) {
|
---|
188 | return trainingPartitions == null || !trainingPartitions.ContainsKey(model) ||
|
---|
189 | (trainingPartitions[model].Start <= currentRow && currentRow < trainingPartitions[model].End);
|
---|
190 | }
|
---|
191 |
|
---|
192 | private bool RowIsTestForModel(int currentRow, IClassificationModel model) {
|
---|
193 | return testPartitions == null || !testPartitions.ContainsKey(model) ||
|
---|
194 | (testPartitions[model].Start <= currentRow && currentRow < testPartitions[model].End);
|
---|
195 | }
|
---|
196 |
|
---|
197 | public override IEnumerable<double> GetEstimatedClassValues(IEnumerable<int> rows) {
|
---|
198 | return from xs in GetEstimatedClassValueVectors(ProblemData.Dataset, rows)
|
---|
199 | select AggregateEstimatedClassValues(xs);
|
---|
200 | }
|
---|
201 |
|
---|
202 | public IEnumerable<IEnumerable<double>> GetEstimatedClassValueVectors(Dataset dataset, IEnumerable<int> rows) {
|
---|
203 | if (!Model.Models.Any()) yield break;
|
---|
204 | var estimatedValuesEnumerators = (from model in Model.Models
|
---|
205 | select model.GetEstimatedClassValues(dataset, rows).GetEnumerator())
|
---|
206 | .ToList();
|
---|
207 |
|
---|
208 | while (estimatedValuesEnumerators.All(en => en.MoveNext())) {
|
---|
209 | yield return from enumerator in estimatedValuesEnumerators
|
---|
210 | select enumerator.Current;
|
---|
211 | }
|
---|
212 | }
|
---|
213 |
|
---|
214 | private double AggregateEstimatedClassValues(IEnumerable<double> estimatedClassValues) {
|
---|
215 | return estimatedClassValues
|
---|
216 | .GroupBy(x => x)
|
---|
217 | .OrderBy(g => -g.Count())
|
---|
218 | .Select(g => g.Key)
|
---|
219 | .DefaultIfEmpty(double.NaN)
|
---|
220 | .First();
|
---|
221 | }
|
---|
222 | #endregion
|
---|
223 |
|
---|
224 | protected override void OnProblemDataChanged() {
|
---|
225 | IClassificationProblemData problemData = new ClassificationProblemData(ProblemData.Dataset,
|
---|
226 | ProblemData.AllowedInputVariables,
|
---|
227 | ProblemData.TargetVariable);
|
---|
228 | problemData.TrainingPartition.Start = ProblemData.TrainingPartition.Start;
|
---|
229 | problemData.TrainingPartition.End = ProblemData.TrainingPartition.End;
|
---|
230 | problemData.TestPartition.Start = ProblemData.TestPartition.Start;
|
---|
231 | problemData.TestPartition.End = ProblemData.TestPartition.End;
|
---|
232 |
|
---|
233 | foreach (var solution in ClassificationSolutions) {
|
---|
234 | if (solution is ClassificationEnsembleSolution)
|
---|
235 | solution.ProblemData = ProblemData;
|
---|
236 | else
|
---|
237 | solution.ProblemData = problemData;
|
---|
238 | }
|
---|
239 | foreach (var trainingPartition in trainingPartitions.Values) {
|
---|
240 | trainingPartition.Start = ProblemData.TrainingPartition.Start;
|
---|
241 | trainingPartition.End = ProblemData.TrainingPartition.End;
|
---|
242 | }
|
---|
243 | foreach (var testPartition in testPartitions.Values) {
|
---|
244 | testPartition.Start = ProblemData.TestPartition.Start;
|
---|
245 | testPartition.End = ProblemData.TestPartition.End;
|
---|
246 | }
|
---|
247 |
|
---|
248 | base.OnProblemDataChanged();
|
---|
249 | }
|
---|
250 |
|
---|
251 | public void AddClassificationSolutions(IEnumerable<IClassificationSolution> solutions) {
|
---|
252 | classificationSolutions.AddRange(solutions);
|
---|
253 | }
|
---|
254 | public void RemoveClassificationSolutions(IEnumerable<IClassificationSolution> solutions) {
|
---|
255 | classificationSolutions.RemoveRange(solutions);
|
---|
256 | }
|
---|
257 |
|
---|
258 | private void classificationSolutions_ItemsAdded(object sender, CollectionItemsChangedEventArgs<IClassificationSolution> e) {
|
---|
259 | foreach (var solution in e.Items) AddClassificationSolution(solution);
|
---|
260 | RecalculateResults();
|
---|
261 | }
|
---|
262 | private void classificationSolutions_ItemsRemoved(object sender, CollectionItemsChangedEventArgs<IClassificationSolution> e) {
|
---|
263 | foreach (var solution in e.Items) RemoveClassificationSolution(solution);
|
---|
264 | RecalculateResults();
|
---|
265 | }
|
---|
266 | private void classificationSolutions_CollectionReset(object sender, CollectionItemsChangedEventArgs<IClassificationSolution> e) {
|
---|
267 | foreach (var solution in e.OldItems) RemoveClassificationSolution(solution);
|
---|
268 | foreach (var solution in e.Items) AddClassificationSolution(solution);
|
---|
269 | RecalculateResults();
|
---|
270 | }
|
---|
271 |
|
---|
272 | private void AddClassificationSolution(IClassificationSolution solution) {
|
---|
273 | if (Model.Models.Contains(solution.Model)) throw new ArgumentException();
|
---|
274 | Model.Add(solution.Model);
|
---|
275 | trainingPartitions[solution.Model] = solution.ProblemData.TrainingPartition;
|
---|
276 | testPartitions[solution.Model] = solution.ProblemData.TestPartition;
|
---|
277 | }
|
---|
278 |
|
---|
279 | private void RemoveClassificationSolution(IClassificationSolution solution) {
|
---|
280 | if (!Model.Models.Contains(solution.Model)) throw new ArgumentException();
|
---|
281 | Model.Remove(solution.Model);
|
---|
282 | trainingPartitions.Remove(solution.Model);
|
---|
283 | testPartitions.Remove(solution.Model);
|
---|
284 | }
|
---|
285 | }
|
---|
286 | }
|
---|