Free cookie consent management tool by TermsFeed Policy Generator

source: branches/3026_IntegrationIntoSymSpace/HeuristicLab.Problems.DataAnalysis.Symbolic.Regression/3.4/SingleObjective/Evaluators/NMSESingleObjectiveConstraintsEvaluator.cs @ 18027

Last change on this file since 18027 was 18027, checked in by dpiringe, 3 years ago

#3026

  • merged trunk into branch
File size: 10.8 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 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
22using System;
23using System.Collections.Generic;
24using System.Linq;
25using HEAL.Attic;
26using HeuristicLab.Common;
27using HeuristicLab.Core;
28using HeuristicLab.Data;
29using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
30using HeuristicLab.Parameters;
31
32namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
33  [Item("NMSE Evaluator with shape-constraints (single-objective)", "Calculates NMSE of a symbolic regression solution and checks constraints. The fitness is a combination of NMSE and constraint violations.")]
34  [StorableType("27473973-DD8D-4375-997D-942E2280AE8E")]
35  public class NMSESingleObjectiveConstraintsEvaluator : SymbolicRegressionSingleObjectiveEvaluator {
36    #region Parameter/Properties
37
38    private const string OptimizeParametersParameterName = "OptimizeParameters";
39    private const string ParameterOptimizationIterationsParameterName = "ParameterOptimizationIterations";
40    private const string UseSoftConstraintsParameterName = "UseSoftConstraintsEvaluation";
41    private const string BoundsEstimatorParameterName = "BoundsEstimator";
42    private const string PenaltyFactorParameterName = "PenaltyFactor";
43
44
45    public IFixedValueParameter<BoolValue> OptimizerParametersParameter =>
46      (IFixedValueParameter<BoolValue>)Parameters[OptimizeParametersParameterName];
47
48    public IFixedValueParameter<IntValue> ConstantOptimizationIterationsParameter =>
49      (IFixedValueParameter<IntValue>)Parameters[ParameterOptimizationIterationsParameterName];
50
51    public IFixedValueParameter<BoolValue> UseSoftConstraintsParameter =>
52      (IFixedValueParameter<BoolValue>)Parameters[UseSoftConstraintsParameterName];
53
54    public IValueParameter<IBoundsEstimator> BoundsEstimatorParameter =>
55      (IValueParameter<IBoundsEstimator>)Parameters[BoundsEstimatorParameterName];
56    public IFixedValueParameter<DoubleValue> PenaltyFactorParameter =>
57      (IFixedValueParameter<DoubleValue>)Parameters[PenaltyFactorParameterName];
58
59    public bool OptimizeParameters {
60      get => OptimizerParametersParameter.Value.Value;
61      set => OptimizerParametersParameter.Value.Value = value;
62    }
63
64    public int ConstantOptimizationIterations {
65      get => ConstantOptimizationIterationsParameter.Value.Value;
66      set => ConstantOptimizationIterationsParameter.Value.Value = value;
67    }
68
69    public bool UseSoftConstraints {
70      get => UseSoftConstraintsParameter.Value.Value;
71      set => UseSoftConstraintsParameter.Value.Value = value;
72    }
73
74    public IBoundsEstimator BoundsEstimator {
75      get => BoundsEstimatorParameter.Value;
76      set => BoundsEstimatorParameter.Value = value;
77    }
78
79    public double PenalityFactor {
80      get => PenaltyFactorParameter.Value.Value;
81      set => PenaltyFactorParameter.Value.Value = value;
82    }
83
84
85    public override bool Maximization => false; // NMSE is minimized
86
87    #endregion
88
89    #region Constructors/Cloning
90
91    [StorableConstructor]
92    protected NMSESingleObjectiveConstraintsEvaluator(StorableConstructorFlag _) : base(_) { }
93
94    protected NMSESingleObjectiveConstraintsEvaluator(
95      NMSESingleObjectiveConstraintsEvaluator original, Cloner cloner) : base(original, cloner) { }
96
97    public NMSESingleObjectiveConstraintsEvaluator() {
98      Parameters.Add(new FixedValueParameter<BoolValue>(OptimizeParametersParameterName,
99        "Define whether optimization of numeric parameters is active or not (default: false).", new BoolValue(false)));
100      Parameters.Add(new FixedValueParameter<IntValue>(ParameterOptimizationIterationsParameterName,
101        "Define how many parameter optimization steps should be performed (default: 10).", new IntValue(10)));
102      Parameters.Add(new FixedValueParameter<BoolValue>(UseSoftConstraintsParameterName,
103        "Define whether the constraints are penalized by soft or hard constraints (default: false).", new BoolValue(false)));
104      Parameters.Add(new ValueParameter<IBoundsEstimator>(BoundsEstimatorParameterName,
105        "The estimator which is used to estimate output ranges of models (default: interval arithmetic).", new IntervalArithBoundsEstimator()));
106      Parameters.Add(new FixedValueParameter<DoubleValue>(PenaltyFactorParameterName,
107        "Punishment factor for constraint violations for soft constraint handling (fitness = NMSE + penaltyFactor * avg(violations)) (default: 1.0)", new DoubleValue(1.0)));
108    }
109
110    [StorableHook(HookType.AfterDeserialization)]
111    private void AfterDeserialization() { }
112
113    public override IDeepCloneable Clone(Cloner cloner) {
114      return new NMSESingleObjectiveConstraintsEvaluator(this, cloner);
115    }
116
117    #endregion
118
119    public override IOperation InstrumentedApply() {
120      var rows = GenerateRowsToEvaluate();
121      var tree = SymbolicExpressionTreeParameter.ActualValue;
122      var problemData = ProblemDataParameter.ActualValue;
123      var interpreter = SymbolicDataAnalysisTreeInterpreterParameter.ActualValue;
124      var estimationLimits = EstimationLimitsParameter.ActualValue;
125      var applyLinearScaling = ApplyLinearScalingParameter.ActualValue.Value;
126
127      if (OptimizeParameters) {
128        SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, tree, problemData, rows,
129          false, ConstantOptimizationIterations, true,
130          estimationLimits.Lower, estimationLimits.Upper);
131      } else {
132        if (applyLinearScaling) {
133          var rootNode = new ProgramRootSymbol().CreateTreeNode();
134          var startNode = new StartSymbol().CreateTreeNode();
135          var offset = tree.Root.GetSubtree(0) //Start
136                                .GetSubtree(0); //Offset
137          var scaling = offset.GetSubtree(0);
138
139          //Check if tree contains offset and scaling nodes
140          if (!(offset.Symbol is Addition) || !(scaling.Symbol is Multiplication))
141            throw new ArgumentException($"{ItemName} can only be used with LinearScalingGrammar.");
142
143          var t = (ISymbolicExpressionTreeNode)scaling.GetSubtree(0).Clone();
144          rootNode.AddSubtree(startNode);
145          startNode.AddSubtree(t);
146          var newTree = new SymbolicExpressionTree(rootNode);
147
148          //calculate alpha and beta for scaling
149          var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(newTree, problemData.Dataset, rows);
150
151          var targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
152          OnlineLinearScalingParameterCalculator.Calculate(estimatedValues, targetValues, out var alpha, out var beta,
153            out var errorState);
154
155          if (errorState == OnlineCalculatorError.None) {
156            //Set alpha and beta to the scaling nodes from ia grammar
157            var offsetParameter = offset.GetSubtree(1) as ConstantTreeNode;
158            offsetParameter.Value = alpha;
159            var scalingParameter = scaling.GetSubtree(1) as ConstantTreeNode;
160            scalingParameter.Value = beta;
161          }
162        } // else: alpha and beta are evolved
163      }
164
165      var quality = Calculate(interpreter, tree, estimationLimits.Lower, estimationLimits.Upper, problemData, rows,
166        BoundsEstimator, UseSoftConstraints, PenalityFactor);
167      QualityParameter.ActualValue = new DoubleValue(quality);
168
169      return base.InstrumentedApply();
170    }
171
172    public static double Calculate(
173      ISymbolicDataAnalysisExpressionTreeInterpreter interpreter,
174      ISymbolicExpressionTree tree,
175      double lowerEstimationLimit, double upperEstimationLimit,
176      IRegressionProblemData problemData, IEnumerable<int> rows,
177      IBoundsEstimator estimator,
178      bool useSoftConstraints = false, double penaltyFactor = 1.0) {
179
180      var estimatedValues = interpreter.GetSymbolicExpressionTreeValues(tree, problemData.Dataset, rows);
181      var targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
182      var constraints = Enumerable.Empty<ShapeConstraint>();
183      if (problemData is ShapeConstrainedRegressionProblemData scProbData) {
184        constraints = scProbData.ShapeConstraints.EnabledConstraints;
185      }
186      var intervalCollection = problemData.VariableRanges;
187
188      var boundedEstimatedValues = estimatedValues.LimitToRange(lowerEstimationLimit, upperEstimationLimit);
189      var nmse = OnlineNormalizedMeanSquaredErrorCalculator.Calculate(targetValues, boundedEstimatedValues,
190        out var errorState);
191
192      if (errorState != OnlineCalculatorError.None) {
193        return 1.0;
194      }
195
196      var constraintViolations = IntervalUtil.GetConstraintViolations(constraints, estimator, intervalCollection, tree);
197
198      if (constraintViolations.Any(x => double.IsNaN(x) || double.IsInfinity(x))) {
199        return 1.0;
200      }
201
202      if (useSoftConstraints) {
203        if (penaltyFactor < 0.0)
204          throw new ArgumentException("The parameter has to be >= 0.0.", nameof(penaltyFactor));
205
206        var weightedViolationsAvg = constraints
207          .Zip(constraintViolations, (c, v) => c.Weight * v)
208          .Average();
209
210        return Math.Min(nmse, 1.0) + penaltyFactor * weightedViolationsAvg;
211      } else if (constraintViolations.Any(x => x > 0.0)) {
212        return 1.0;
213      }
214
215      return nmse;
216    }
217
218    public override double Evaluate(
219      IExecutionContext context, ISymbolicExpressionTree tree, IRegressionProblemData problemData,
220      IEnumerable<int> rows) {
221      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = context;
222      EstimationLimitsParameter.ExecutionContext = context;
223      ApplyLinearScalingParameter.ExecutionContext = context;
224
225      var nmse = Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, tree,
226        EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper,
227        problemData, rows, BoundsEstimator, UseSoftConstraints, PenalityFactor);
228
229      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = null;
230      EstimationLimitsParameter.ExecutionContext = null;
231      ApplyLinearScalingParameter.ExecutionContext = null;
232
233      return nmse;
234    }
235  }
236}
Note: See TracBrowser for help on using the repository browser.