1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2016 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.Encodings.SymbolicExpressionTreeEncoding;
|
---|
29 | using HeuristicLab.Optimization;
|
---|
30 | using HeuristicLab.Parameters;
|
---|
31 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
32 |
|
---|
33 | namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
|
---|
34 | [Item("SymbolicRegressionSingleObjectiveOSGAEvaluator", "An evaluator which tries to predict when a child will not be able to fullfil offspring selection criteria, to save evaluation time.")]
|
---|
35 | [StorableClass]
|
---|
36 | public class SymbolicRegressionSingleObjectiveOsgaEvaluator : SymbolicRegressionSingleObjectiveEvaluator {
|
---|
37 | private const string RelativeParentChildQualityThresholdParameterName = "RelativeParentChildQualityThreshold";
|
---|
38 | private const string RelativeFitnessEvaluationIntervalSizeParameterName = "RelativeFitnessEvaluationIntervalSize";
|
---|
39 | private const string ResultCollectionParameterName = "Results";
|
---|
40 | private const string AggregateStatisticsParameterName = "AggregateStatistics";
|
---|
41 | private const string ActualSelectionPressureParameterName = "SelectionPressure";
|
---|
42 | private const string UseAdaptiveQualityThresholdParameterName = "UseAdaptiveQualityThreshold";
|
---|
43 | private const string UseFixedEvaluationIntervalsParameterName = "UseFixedEvaluationIntervals";
|
---|
44 |
|
---|
45 | #region parameters
|
---|
46 | public IFixedValueParameter<BoolValue> UseFixedEvaluationIntervalsParameter {
|
---|
47 | get { return (IFixedValueParameter<BoolValue>)Parameters[UseFixedEvaluationIntervalsParameterName]; }
|
---|
48 | }
|
---|
49 | public IFixedValueParameter<BoolValue> UseAdaptiveQualityThresholdParameter {
|
---|
50 | get { return (IFixedValueParameter<BoolValue>)Parameters[UseAdaptiveQualityThresholdParameterName]; }
|
---|
51 | }
|
---|
52 | public ILookupParameter<DoubleValue> ActualSelectionPressureParameter {
|
---|
53 | get { return (ILookupParameter<DoubleValue>)Parameters[ActualSelectionPressureParameterName]; }
|
---|
54 | }
|
---|
55 | public ILookupParameter<ResultCollection> ResultCollectionParameter {
|
---|
56 | get { return (ILookupParameter<ResultCollection>)Parameters[ResultCollectionParameterName]; }
|
---|
57 | }
|
---|
58 | public IValueParameter<BoolValue> AggregateStatisticsParameter {
|
---|
59 | get { return (IValueParameter<BoolValue>)Parameters[AggregateStatisticsParameterName]; }
|
---|
60 | }
|
---|
61 | public IValueParameter<IntMatrix> RejectedStatsParameter {
|
---|
62 | get { return (IValueParameter<IntMatrix>)Parameters["RejectedStats"]; }
|
---|
63 | }
|
---|
64 | public IValueParameter<IntMatrix> NotRejectedStatsParameter {
|
---|
65 | get { return (IValueParameter<IntMatrix>)Parameters["TotalStats"]; }
|
---|
66 | }
|
---|
67 | public IValueLookupParameter<DoubleValue> ComparisonFactorParameter {
|
---|
68 | get { return (ValueLookupParameter<DoubleValue>)Parameters["ComparisonFactor"]; }
|
---|
69 | }
|
---|
70 | public IFixedValueParameter<PercentValue> RelativeParentChildQualityThresholdParameter {
|
---|
71 | get { return (IFixedValueParameter<PercentValue>)Parameters[RelativeParentChildQualityThresholdParameterName]; }
|
---|
72 | }
|
---|
73 | public IFixedValueParameter<PercentValue> RelativeFitnessEvaluationIntervalSizeParameter {
|
---|
74 | get { return (IFixedValueParameter<PercentValue>)Parameters[RelativeFitnessEvaluationIntervalSizeParameterName]; }
|
---|
75 | }
|
---|
76 | public IScopeTreeLookupParameter<DoubleValue> ParentQualitiesParameter { get { return (IScopeTreeLookupParameter<DoubleValue>)Parameters["ParentQualities"]; } }
|
---|
77 | #endregion
|
---|
78 |
|
---|
79 | #region parameter properties
|
---|
80 | public bool UseFixedEvaluationIntervals {
|
---|
81 | get { return UseFixedEvaluationIntervalsParameter.Value.Value; }
|
---|
82 | set { UseFixedEvaluationIntervalsParameter.Value.Value = value; }
|
---|
83 | }
|
---|
84 | public bool UseAdaptiveQualityThreshold {
|
---|
85 | get { return UseAdaptiveQualityThresholdParameter.Value.Value; }
|
---|
86 | set { UseAdaptiveQualityThresholdParameter.Value.Value = value; }
|
---|
87 | }
|
---|
88 | public double RelativeParentChildQualityThreshold {
|
---|
89 | get { return RelativeParentChildQualityThresholdParameter.Value.Value; }
|
---|
90 | set { RelativeParentChildQualityThresholdParameter.Value.Value = value; }
|
---|
91 | }
|
---|
92 |
|
---|
93 | public double RelativeFitnessEvaluationIntervalSize {
|
---|
94 | get { return RelativeFitnessEvaluationIntervalSizeParameter.Value.Value; }
|
---|
95 | set { RelativeFitnessEvaluationIntervalSizeParameter.Value.Value = value; }
|
---|
96 | }
|
---|
97 |
|
---|
98 | public IntMatrix RejectedStats {
|
---|
99 | get { return RejectedStatsParameter.Value; }
|
---|
100 | set { RejectedStatsParameter.Value = value; }
|
---|
101 | }
|
---|
102 |
|
---|
103 | public IntMatrix TotalStats {
|
---|
104 | get { return NotRejectedStatsParameter.Value; }
|
---|
105 | set { NotRejectedStatsParameter.Value = value; }
|
---|
106 | }
|
---|
107 | #endregion
|
---|
108 |
|
---|
109 | public override bool Maximization {
|
---|
110 | get { return true; }
|
---|
111 | }
|
---|
112 |
|
---|
113 | public SymbolicRegressionSingleObjectiveOsgaEvaluator() {
|
---|
114 | Parameters.Add(new ValueLookupParameter<DoubleValue>("ComparisonFactor", "Determines if the quality should be compared to the better parent (1.0), to the worse (0.0) or to any linearly interpolated value between them."));
|
---|
115 | Parameters.Add(new FixedValueParameter<PercentValue>(RelativeParentChildQualityThresholdParameterName, new PercentValue(0.9)));
|
---|
116 | Parameters.Add(new FixedValueParameter<PercentValue>(RelativeFitnessEvaluationIntervalSizeParameterName, new PercentValue(0.1)));
|
---|
117 | Parameters.Add(new LookupParameter<ResultCollection>(ResultCollectionParameterName));
|
---|
118 | Parameters.Add(new ScopeTreeLookupParameter<DoubleValue>("ParentQualities") { ActualName = "Quality" });
|
---|
119 | Parameters.Add(new ValueParameter<IntMatrix>("RejectedStats", new IntMatrix()));
|
---|
120 | Parameters.Add(new ValueParameter<IntMatrix>("TotalStats", new IntMatrix()));
|
---|
121 | Parameters.Add(new ValueParameter<BoolValue>(AggregateStatisticsParameterName, new BoolValue(false)));
|
---|
122 | Parameters.Add(new LookupParameter<DoubleValue>(ActualSelectionPressureParameterName));
|
---|
123 | Parameters.Add(new FixedValueParameter<BoolValue>(UseAdaptiveQualityThresholdParameterName, new BoolValue(false)));
|
---|
124 | Parameters.Add(new FixedValueParameter<BoolValue>(UseFixedEvaluationIntervalsParameterName, new BoolValue(false)));
|
---|
125 | }
|
---|
126 |
|
---|
127 | [StorableHook(HookType.AfterDeserialization)]
|
---|
128 | private void AfterDeserialization() {
|
---|
129 | if (!Parameters.ContainsKey(ActualSelectionPressureParameterName))
|
---|
130 | Parameters.Add(new LookupParameter<DoubleValue>(ActualSelectionPressureParameterName));
|
---|
131 |
|
---|
132 | if (!Parameters.ContainsKey(UseAdaptiveQualityThresholdParameterName))
|
---|
133 | Parameters.Add(new FixedValueParameter<BoolValue>(UseAdaptiveQualityThresholdParameterName, new BoolValue(false)));
|
---|
134 |
|
---|
135 | if (!Parameters.ContainsKey(UseFixedEvaluationIntervalsParameterName))
|
---|
136 | Parameters.Add(new FixedValueParameter<BoolValue>(UseFixedEvaluationIntervalsParameterName, new BoolValue(false)));
|
---|
137 | }
|
---|
138 |
|
---|
139 | [StorableConstructor]
|
---|
140 | protected SymbolicRegressionSingleObjectiveOsgaEvaluator(bool deserializing) : base(deserializing) { }
|
---|
141 |
|
---|
142 | protected SymbolicRegressionSingleObjectiveOsgaEvaluator(SymbolicRegressionSingleObjectiveOsgaEvaluator original, Cloner cloner) : base(original, cloner) { }
|
---|
143 |
|
---|
144 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
145 | return new SymbolicRegressionSingleObjectiveOsgaEvaluator(this, cloner);
|
---|
146 | }
|
---|
147 |
|
---|
148 | public override void ClearState() {
|
---|
149 | base.ClearState();
|
---|
150 | RejectedStats = new IntMatrix();
|
---|
151 | TotalStats = new IntMatrix();
|
---|
152 | }
|
---|
153 |
|
---|
154 | public override IOperation InstrumentedApply() {
|
---|
155 | var solution = SymbolicExpressionTreeParameter.ActualValue;
|
---|
156 | IEnumerable<int> rows = GenerateRowsToEvaluate();
|
---|
157 |
|
---|
158 | var interpreter = SymbolicDataAnalysisTreeInterpreterParameter.ActualValue;
|
---|
159 | var estimationLimits = EstimationLimitsParameter.ActualValue;
|
---|
160 | var problemData = ProblemDataParameter.ActualValue;
|
---|
161 | var applyLinearScaling = ApplyLinearScalingParameter.ActualValue.Value;
|
---|
162 |
|
---|
163 | double quality;
|
---|
164 | var parentQualities = ParentQualitiesParameter.ActualValue;
|
---|
165 |
|
---|
166 | // parent subscopes are not present during evaluation of the initial population
|
---|
167 | if (parentQualities.Length > 0) {
|
---|
168 | quality = Calculate(interpreter, solution, estimationLimits, problemData, rows);
|
---|
169 | } else {
|
---|
170 | quality = Calculate(interpreter, solution, estimationLimits.Lower, estimationLimits.Upper, problemData, rows, applyLinearScaling);
|
---|
171 | }
|
---|
172 | QualityParameter.ActualValue = new DoubleValue(quality);
|
---|
173 |
|
---|
174 | return base.InstrumentedApply();
|
---|
175 | }
|
---|
176 |
|
---|
177 | public static double Calculate(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, ISymbolicExpressionTree solution, double lowerEstimationLimit, double upperEstimationLimit, IRegressionProblemData problemData, IEnumerable<int> rows, bool applyLinearScaling) {
|
---|
178 | IEnumerable<double> estimatedValues = interpreter.GetSymbolicExpressionTreeValues(solution, problemData.Dataset, rows);
|
---|
179 | IEnumerable<double> targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
|
---|
180 | OnlineCalculatorError errorState;
|
---|
181 |
|
---|
182 | double r;
|
---|
183 | if (applyLinearScaling) {
|
---|
184 | var rCalculator = new OnlinePearsonsRCalculator();
|
---|
185 | CalculateWithScaling(targetValues, estimatedValues, lowerEstimationLimit, upperEstimationLimit, rCalculator, problemData.Dataset.Rows);
|
---|
186 | errorState = rCalculator.ErrorState;
|
---|
187 | r = rCalculator.R;
|
---|
188 | } else {
|
---|
189 | IEnumerable<double> boundedEstimatedValues = estimatedValues.LimitToRange(lowerEstimationLimit, upperEstimationLimit);
|
---|
190 | r = OnlinePearsonsRCalculator.Calculate(targetValues, boundedEstimatedValues, out errorState);
|
---|
191 | }
|
---|
192 | if (errorState != OnlineCalculatorError.None) return double.NaN;
|
---|
193 | return r * r;
|
---|
194 | }
|
---|
195 |
|
---|
196 | private double Calculate(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, ISymbolicExpressionTree solution, DoubleLimit estimationLimits, IRegressionProblemData problemData, IEnumerable<int> rows) {
|
---|
197 | var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(solution, problemData.Dataset, rows).LimitToRange(estimationLimits.Lower, estimationLimits.Upper);
|
---|
198 | var targetValues = problemData.Dataset.GetReadOnlyDoubleValues(problemData.TargetVariable);
|
---|
199 |
|
---|
200 | var parentQualities = ParentQualitiesParameter.ActualValue.Select(x => x.Value);
|
---|
201 | var minQuality = parentQualities.Min();
|
---|
202 | var maxQuality = parentQualities.Max();
|
---|
203 | var comparisonFactor = ComparisonFactorParameter.ActualValue.Value;
|
---|
204 | var parentQuality = minQuality + (maxQuality - minQuality) * comparisonFactor;
|
---|
205 |
|
---|
206 | var e = estimatedValues.GetEnumerator();
|
---|
207 |
|
---|
208 | #region fixed intervals
|
---|
209 | if (UseFixedEvaluationIntervals) {
|
---|
210 | double threshold = parentQuality * RelativeParentChildQualityThreshold;
|
---|
211 | if (UseAdaptiveQualityThreshold) {
|
---|
212 | var actualSelectionPressure = ActualSelectionPressureParameter.ActualValue;
|
---|
213 | if (actualSelectionPressure != null)
|
---|
214 | threshold = parentQuality * (1 - actualSelectionPressure.Value / 100.0);
|
---|
215 | }
|
---|
216 |
|
---|
217 | var pearsonRCalculator = new OnlinePearsonsRCalculator();
|
---|
218 | var targetValuesEnumerator = targetValues.GetEnumerator();
|
---|
219 | var trainingPartitionSize = problemData.TrainingPartition.Size;
|
---|
220 | var interval = (int)Math.Floor(trainingPartitionSize * RelativeFitnessEvaluationIntervalSize);
|
---|
221 |
|
---|
222 | var aggregateStatistics = AggregateStatisticsParameter.Value.Value;
|
---|
223 | var i = 0;
|
---|
224 | if (aggregateStatistics) {
|
---|
225 | var trainingEnd = problemData.TrainingPartition.End;
|
---|
226 | var qualityPerInterval = new List<double>();
|
---|
227 | while (targetValuesEnumerator.MoveNext() && e.MoveNext()) {
|
---|
228 | pearsonRCalculator.Add(targetValuesEnumerator.Current, e.Current);
|
---|
229 | ++i;
|
---|
230 | if (i % interval == 0 || i == trainingPartitionSize) {
|
---|
231 | var q = pearsonRCalculator.ErrorState != OnlineCalculatorError.None ? double.NaN : pearsonRCalculator.R;
|
---|
232 | qualityPerInterval.Add(q * q);
|
---|
233 | }
|
---|
234 | }
|
---|
235 | var r = pearsonRCalculator.ErrorState != OnlineCalculatorError.None ? double.NaN : pearsonRCalculator.R;
|
---|
236 | var actualQuality = r * r;
|
---|
237 |
|
---|
238 | bool predictedRejected = false;
|
---|
239 |
|
---|
240 | i = 0;
|
---|
241 | double quality = actualQuality;
|
---|
242 | foreach (var q in qualityPerInterval) {
|
---|
243 | if (double.IsNaN(q) || !(q > threshold)) {
|
---|
244 | predictedRejected = true;
|
---|
245 | quality = q;
|
---|
246 | break;
|
---|
247 | }
|
---|
248 | ++i;
|
---|
249 | }
|
---|
250 |
|
---|
251 | var actuallyRejected = !(actualQuality > parentQuality);
|
---|
252 |
|
---|
253 | if (RejectedStats.Rows == 0 || TotalStats.Rows == 0) {
|
---|
254 | RejectedStats = new IntMatrix(2, qualityPerInterval.Count);
|
---|
255 | RejectedStats.RowNames = new[] { "Predicted", "Actual" };
|
---|
256 | RejectedStats.ColumnNames = Enumerable.Range(1, RejectedStats.Columns).Select(x => string.Format("0-{0}", Math.Min(trainingEnd, x * interval)));
|
---|
257 | TotalStats = new IntMatrix(2, 2);
|
---|
258 | TotalStats.RowNames = new[] { "Predicted", "Actual" };
|
---|
259 | TotalStats.ColumnNames = new[] { "Rejected", "Not Rejected" };
|
---|
260 | }
|
---|
261 | // gather some statistics
|
---|
262 | if (predictedRejected) {
|
---|
263 | RejectedStats[0, i]++;
|
---|
264 | TotalStats[0, 0]++;
|
---|
265 | } else {
|
---|
266 | TotalStats[0, 1]++;
|
---|
267 | }
|
---|
268 | if (actuallyRejected) {
|
---|
269 | TotalStats[1, 0]++;
|
---|
270 | } else {
|
---|
271 | TotalStats[1, 1]++;
|
---|
272 | }
|
---|
273 | if (predictedRejected && actuallyRejected) {
|
---|
274 | RejectedStats[1, i]++;
|
---|
275 | }
|
---|
276 | return quality;
|
---|
277 | } else {
|
---|
278 | while (targetValuesEnumerator.MoveNext() && e.MoveNext()) {
|
---|
279 | pearsonRCalculator.Add(targetValuesEnumerator.Current, e.Current);
|
---|
280 | ++i;
|
---|
281 | if (i % interval == 0 || i == trainingPartitionSize) {
|
---|
282 | var q = pearsonRCalculator.ErrorState != OnlineCalculatorError.None ? double.NaN : pearsonRCalculator.R;
|
---|
283 | var quality = q * q;
|
---|
284 | if (!(quality > threshold))
|
---|
285 | return quality;
|
---|
286 | }
|
---|
287 | }
|
---|
288 | var r = pearsonRCalculator.ErrorState != OnlineCalculatorError.None ? double.NaN : pearsonRCalculator.R;
|
---|
289 | var actualQuality = r * r;
|
---|
290 | return actualQuality;
|
---|
291 | }
|
---|
292 | #endregion
|
---|
293 | } else {
|
---|
294 | var calculator = new OnlinePearsonsRCalculator();
|
---|
295 | var trainingPartitionSize = problemData.TrainingPartition.Size;
|
---|
296 | var interval = (int)Math.Floor(trainingPartitionSize * RelativeFitnessEvaluationIntervalSize);
|
---|
297 | double quality = double.NaN;
|
---|
298 | var estimated = new List<double>(); // save estimated values in a list so we don't re-evaluate
|
---|
299 | // use the actual estimated values for the first i * interval rows of the training partition and and assume the remaining rows are perfectly correlated
|
---|
300 | // if the quality of the individual still falls below the parent quality, then we can reject it sooner, otherwise as i increases the whole estimated series will be used
|
---|
301 | for (int i = 0; i < trainingPartitionSize; i += interval) {
|
---|
302 | calculator.Reset();
|
---|
303 | // save estimated values into the list (for caching)
|
---|
304 | int j = i;
|
---|
305 | int end = Math.Min(trainingPartitionSize, i + interval);
|
---|
306 | while (j < end && e.MoveNext()) {
|
---|
307 | estimated.Add(e.Current);
|
---|
308 | j++;
|
---|
309 | }
|
---|
310 | var start = problemData.TrainingPartition.Start;
|
---|
311 | // add (estimated, target) pairs to the calculator
|
---|
312 | for (j = 0; j < end; ++j) {
|
---|
313 | var index = j + start;
|
---|
314 | calculator.Add(targetValues[index], estimated[j]);
|
---|
315 | }
|
---|
316 | // add (target, target) pairs to the calculator (simulate perfect correlation on the remaining rows)
|
---|
317 | for (; j < trainingPartitionSize; ++j) {
|
---|
318 | var index = j + start;
|
---|
319 | var v = targetValues[index];
|
---|
320 | calculator.Add(v, v);
|
---|
321 | }
|
---|
322 | var r = calculator.ErrorState == OnlineCalculatorError.None ? calculator.R : double.NaN;
|
---|
323 | quality = r * r;
|
---|
324 | if (!(quality > parentQuality))
|
---|
325 | break;
|
---|
326 | }
|
---|
327 | return quality;
|
---|
328 | }
|
---|
329 | }
|
---|
330 |
|
---|
331 | public override double Evaluate(IExecutionContext context, ISymbolicExpressionTree tree, IRegressionProblemData problemData, IEnumerable<int> rows) {
|
---|
332 | SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = context;
|
---|
333 | EstimationLimitsParameter.ExecutionContext = context;
|
---|
334 | ApplyLinearScalingParameter.ExecutionContext = context;
|
---|
335 |
|
---|
336 | var interpreter = SymbolicDataAnalysisTreeInterpreterParameter.ActualValue;
|
---|
337 | var estimationLimits = EstimationLimitsParameter.ActualValue;
|
---|
338 | var applyLinearScaling = ApplyLinearScalingParameter.ActualValue.Value;
|
---|
339 |
|
---|
340 | double r2 = Calculate(interpreter, tree, estimationLimits.Lower, estimationLimits.Upper, problemData, rows, applyLinearScaling);
|
---|
341 |
|
---|
342 | SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = null;
|
---|
343 | EstimationLimitsParameter.ExecutionContext = null;
|
---|
344 | ApplyLinearScalingParameter.ExecutionContext = null;
|
---|
345 |
|
---|
346 | return r2;
|
---|
347 | }
|
---|
348 | }
|
---|
349 | }
|
---|