1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2012 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.Common;
|
---|
26 | using HeuristicLab.Core;
|
---|
27 | using HeuristicLab.Data;
|
---|
28 | using HeuristicLab.Optimization;
|
---|
29 | using HeuristicLab.Parameters;
|
---|
30 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
31 | using HeuristicLab.Problems.DataAnalysis;
|
---|
32 | using HeuristicLab.Random;
|
---|
33 |
|
---|
34 | namespace HeuristicLab.Algorithms.DataAnalysis {
|
---|
35 | /// <summary>
|
---|
36 | /// 1R classification algorithm.
|
---|
37 | /// </summary>
|
---|
38 | [Item("OneR", "1R classification algorithm.")]
|
---|
39 | [Creatable("Data Analysis")]
|
---|
40 | [StorableClass]
|
---|
41 | public sealed class OneR : FixedDataAnalysisAlgorithm<IClassificationProblem> {
|
---|
42 |
|
---|
43 | public IValueParameter<IntValue> MinBucketSizeParameter {
|
---|
44 | get { return (IValueParameter<IntValue>)Parameters["MinBucketSize"]; }
|
---|
45 | }
|
---|
46 | public IValueParameter<IRandom> RandomParameter {
|
---|
47 | get { return (IValueParameter<IRandom>)Parameters["Random"]; }
|
---|
48 | }
|
---|
49 |
|
---|
50 | [StorableConstructor]
|
---|
51 | private OneR(bool deserializing) : base(deserializing) { }
|
---|
52 | private OneR(OneR original, Cloner cloner)
|
---|
53 | : base(original, cloner) {
|
---|
54 | }
|
---|
55 | public OneR()
|
---|
56 | : base() {
|
---|
57 | Parameters.Add(new ValueParameter<IntValue>("MinBucketSize", "Minimum size of a bucket for numerical values. (Except for the rightmost bucket)", new IntValue(6)));
|
---|
58 | Parameters.Add(new ValueParameter<IRandom>("Random", "Random number generator", new FastRandom()));
|
---|
59 | Problem = new ClassificationProblem();
|
---|
60 | }
|
---|
61 |
|
---|
62 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
63 | return new OneR(this, cloner);
|
---|
64 | }
|
---|
65 |
|
---|
66 | protected override void Run() {
|
---|
67 | var startTime = DateTime.Now;
|
---|
68 | var solution = CreateOneRSolution(Problem.ProblemData, MinBucketSizeParameter.Value.Value, RandomParameter.Value);
|
---|
69 | Results.Add(new Result("OneR solution", "The 1R classifier.", solution));
|
---|
70 | Results.Add(new Result("OneR Execution Time", "", new TimeSpanValue(DateTime.Now - startTime)));
|
---|
71 |
|
---|
72 | startTime = DateTime.Now;
|
---|
73 | var solution3 = OneRTest.CreateOneRSolution(Problem.ProblemData, MinBucketSizeParameter.Value.Value);
|
---|
74 | Results.Add(new Result("OneR Test2 solution", "The 1R classifier.", solution3));
|
---|
75 | Results.Add(new Result("OneR Test2 Execution", "", new TimeSpanValue(DateTime.Now - startTime)));
|
---|
76 | }
|
---|
77 |
|
---|
78 | public static IClassificationSolution CreateOneRSolution(IClassificationProblemData problemData, int minBucketSize, IRandom random) {
|
---|
79 | Dataset dataset = problemData.Dataset;
|
---|
80 | var trainingIndices = problemData.TrainingIndices;
|
---|
81 | int rowCount = trainingIndices.Count();
|
---|
82 | string target = problemData.TargetVariable;
|
---|
83 | var inputVariables = problemData.AllowedInputVariables.ToArray();
|
---|
84 | var classValues = problemData.ClassValues.ToArray();
|
---|
85 | double dominatingClass;
|
---|
86 |
|
---|
87 | string bestVariable = null;
|
---|
88 | Dictionary<double, double> bestSplits = null;
|
---|
89 | double missingValuesClass = double.NaN;
|
---|
90 | int correctClassified = 0;
|
---|
91 |
|
---|
92 | for (int variable = 0; variable < inputVariables.Length; variable++) {
|
---|
93 | var inputVariableValues = dataset.GetDoubleValues(inputVariables[variable], trainingIndices).ToArray();
|
---|
94 | var classValuesInDataset = dataset.GetDoubleValues(target, trainingIndices).ToArray();
|
---|
95 |
|
---|
96 | int curCorrectClassified = 0;
|
---|
97 | Dictionary<double, int> classCount = PrepareClassCountDictionary(classValues);
|
---|
98 | Array.Sort(inputVariableValues, classValuesInDataset);
|
---|
99 | double curSplit = Double.NegativeInfinity;
|
---|
100 | Dictionary<double, double> splits = new Dictionary<double, double>();
|
---|
101 | bool newBucket = true;
|
---|
102 | bool done = false;
|
---|
103 | int curRow = 0;
|
---|
104 |
|
---|
105 | if (curRow < inputVariableValues.Length && Double.IsNaN(inputVariableValues[curRow])) {
|
---|
106 | while (curRow < inputVariableValues.Length && Double.IsNaN(inputVariableValues[curRow])) {
|
---|
107 | classCount[classValuesInDataset[curRow]] += 1;
|
---|
108 | curRow++;
|
---|
109 | }
|
---|
110 | if (ExistsDominatingClass(classCount, out dominatingClass)) {
|
---|
111 | missingValuesClass = dominatingClass;
|
---|
112 | } else {
|
---|
113 | missingValuesClass = GetRandomMaxClass(classCount, random);
|
---|
114 | }
|
---|
115 | correctClassified += classCount[missingValuesClass];
|
---|
116 | classCount = PrepareClassCountDictionary(classValues);
|
---|
117 | }
|
---|
118 | while (curRow < inputVariableValues.Length) {
|
---|
119 | if (newBucket) {
|
---|
120 | for (int i = 0; i < minBucketSize && curRow + i < inputVariableValues.Length; i++) {
|
---|
121 | classCount[classValuesInDataset[curRow + i]] += 1;
|
---|
122 | }
|
---|
123 | curRow += minBucketSize;
|
---|
124 | if (curRow >= inputVariableValues.Length) {
|
---|
125 | break;
|
---|
126 | }
|
---|
127 | curSplit = inputVariableValues[curRow];
|
---|
128 | curRow = SetCurRowToEndOfSplit(curRow, inputVariableValues, classValuesInDataset, classCount, curSplit);
|
---|
129 | newBucket = false;
|
---|
130 | }
|
---|
131 |
|
---|
132 | if (ExistsDominatingClass(classCount, out dominatingClass)) {
|
---|
133 | while (curRow + 1 < classValuesInDataset.Length &&
|
---|
134 | IsNextSplitStillDominatingClass(curRow, inputVariableValues, classValuesInDataset, curSplit, dominatingClass)) {
|
---|
135 | curRow++;
|
---|
136 | curSplit = inputVariableValues[curRow];
|
---|
137 | classCount[classValuesInDataset[curRow]] += 1;
|
---|
138 | curRow = SetCurRowToEndOfSplit(curRow, inputVariableValues, classValuesInDataset, classCount, curSplit);
|
---|
139 | }
|
---|
140 |
|
---|
141 | curCorrectClassified += classCount[dominatingClass];
|
---|
142 | done = curRow >= inputVariableValues.Length - 1;
|
---|
143 |
|
---|
144 | if (done) {
|
---|
145 | curSplit = Double.PositiveInfinity;
|
---|
146 | splits.Add(curSplit, dominatingClass);
|
---|
147 | break;
|
---|
148 | }
|
---|
149 |
|
---|
150 | curRow++;
|
---|
151 | //intervals exclude end
|
---|
152 | curSplit = inputVariableValues[curRow];
|
---|
153 | splits.Add(curSplit, dominatingClass);
|
---|
154 |
|
---|
155 | //intervals include start
|
---|
156 | curSplit = inputVariableValues[curRow];
|
---|
157 | classCount = PrepareClassCountDictionary(classValues);
|
---|
158 | newBucket = true;
|
---|
159 | } else {
|
---|
160 | curSplit = inputVariableValues[curRow];
|
---|
161 | classCount[classValuesInDataset[curRow]] += 1;
|
---|
162 | curRow = SetCurRowToEndOfSplit(curRow, inputVariableValues, classValuesInDataset, classCount, curSplit);
|
---|
163 | }
|
---|
164 | }
|
---|
165 |
|
---|
166 | if (!done) {
|
---|
167 | curSplit = Double.PositiveInfinity;
|
---|
168 | double randomClass = GetRandomMaxClass(classCount, random);
|
---|
169 | splits.Add(curSplit, randomClass);
|
---|
170 |
|
---|
171 | curCorrectClassified += classCount[randomClass];
|
---|
172 | }
|
---|
173 |
|
---|
174 | if (curCorrectClassified > correctClassified) {
|
---|
175 | bestVariable = inputVariables[variable];
|
---|
176 | bestSplits = splits;
|
---|
177 | }
|
---|
178 | }
|
---|
179 |
|
---|
180 | //merge intervals to simplify symbolic expression tree
|
---|
181 | Dictionary<double, double> mergedSplits = MergeSplits(bestSplits);
|
---|
182 |
|
---|
183 | var model = new OneRClassificationModel(bestVariable, mergedSplits.Keys.ToArray(), mergedSplits.Values.ToArray(), missingValuesClass);
|
---|
184 | var solution = new OneRClassificationSolution(model, (IClassificationProblemData)problemData.Clone());
|
---|
185 |
|
---|
186 | return solution;
|
---|
187 | }
|
---|
188 |
|
---|
189 | private static double GetRandomMaxClass(Dictionary<double, int> classCount, IRandom random) {
|
---|
190 | IList<double> possibleClasses = new List<double>();
|
---|
191 | int max = 0;
|
---|
192 | foreach (var item in classCount) {
|
---|
193 | if (max < item.Value) {
|
---|
194 | max = item.Value;
|
---|
195 | possibleClasses = new List<double>();
|
---|
196 | possibleClasses.Add(item.Key);
|
---|
197 | } else if (max == item.Value) {
|
---|
198 | possibleClasses.Add(item.Key);
|
---|
199 | }
|
---|
200 | }
|
---|
201 | int classindex = random.Next(possibleClasses.Count);
|
---|
202 | return possibleClasses[classindex];
|
---|
203 | }
|
---|
204 |
|
---|
205 | private static bool IsNextSplitStillDominatingClass(int curRow, double[] inputVariableValues, double[] classValuesInDataset, double curSplit, double dominatingClass) {
|
---|
206 | if (curRow >= classValuesInDataset.Length) {
|
---|
207 | return false;
|
---|
208 | }
|
---|
209 | double nextSplit = inputVariableValues[curRow + 1];
|
---|
210 | int i = 1;
|
---|
211 | while (curRow + i < classValuesInDataset.Length
|
---|
212 | && inputVariableValues[curRow + i] == nextSplit
|
---|
213 | && classValuesInDataset[curRow + i] == dominatingClass) {
|
---|
214 | i++;
|
---|
215 | }
|
---|
216 | if (curRow + i >= classValuesInDataset.Length) {
|
---|
217 | return true;
|
---|
218 | }
|
---|
219 | if (inputVariableValues[curRow + i] != nextSplit) {
|
---|
220 | return true;
|
---|
221 | }
|
---|
222 | // the next split would also contain values of a class which
|
---|
223 | // is not dominating (classValuesInDataset[curRow + i] != dominatingClass)
|
---|
224 | return false;
|
---|
225 | }
|
---|
226 |
|
---|
227 | // needed if variable contains the same value several times
|
---|
228 | private static int SetCurRowToEndOfSplit(int curRow, double[] inputVariableValues, double[] classValuesInDataset, Dictionary<double, int> classCount, double curSplit) {
|
---|
229 | while (curRow + 1 < inputVariableValues.Length && inputVariableValues[curRow + 1] == curSplit) {
|
---|
230 | curRow++;
|
---|
231 | classCount[classValuesInDataset[curRow]] += 1;
|
---|
232 | }
|
---|
233 | return curRow;
|
---|
234 | }
|
---|
235 |
|
---|
236 | private static Dictionary<double, double> MergeSplits(Dictionary<double, double> bestSplits) {
|
---|
237 | Dictionary<double, double> mergedSplits = new Dictionary<double, double>();
|
---|
238 | double nextSplit, nextClass;
|
---|
239 | nextSplit = nextClass = double.NaN;
|
---|
240 | foreach (var item in bestSplits) {
|
---|
241 | if (Double.IsNaN(nextSplit)) {
|
---|
242 | nextSplit = item.Key;
|
---|
243 | nextClass = item.Value;
|
---|
244 | } else {
|
---|
245 | if (nextClass == item.Value) {
|
---|
246 | nextSplit = item.Key;
|
---|
247 | } else {
|
---|
248 | mergedSplits.Add(nextSplit, nextClass);
|
---|
249 | nextSplit = item.Key;
|
---|
250 | nextClass = item.Value;
|
---|
251 | }
|
---|
252 | }
|
---|
253 | }
|
---|
254 | mergedSplits.Add(nextSplit, nextClass);
|
---|
255 | return mergedSplits;
|
---|
256 | }
|
---|
257 |
|
---|
258 | private static bool ExistsDominatingClass(Dictionary<double, int> classCount, out double dominatingClass) {
|
---|
259 | bool dominating = false;
|
---|
260 | int count = 0;
|
---|
261 | dominatingClass = double.NaN;
|
---|
262 | foreach (var item in classCount) {
|
---|
263 | if (item.Value > count) {
|
---|
264 | dominatingClass = item.Key;
|
---|
265 | count = item.Value;
|
---|
266 | dominating = true;
|
---|
267 | } else if (item.Value == count) {
|
---|
268 | dominatingClass = double.NaN;
|
---|
269 | dominating = false;
|
---|
270 | }
|
---|
271 | }
|
---|
272 | return dominating;
|
---|
273 | }
|
---|
274 |
|
---|
275 | private static Dictionary<double, int> PrepareClassCountDictionary(double[] classValues) {
|
---|
276 | Dictionary<double, int> classCount = new Dictionary<double, int>();
|
---|
277 | for (int i = 0; i < classValues.Length; i++) {
|
---|
278 | classCount[classValues[i]] = 0;
|
---|
279 | }
|
---|
280 | return classCount;
|
---|
281 | }
|
---|
282 | }
|
---|
283 | }
|
---|