Free cookie consent management tool by TermsFeed Policy Generator

source: branches/2435-alglib_3_15/HeuristicLab.Problems.DataAnalysis.Symbolic.Regression/3.4/SingleObjective/Evaluators/SymbolicRegressionConstantOptimizationEvaluator.cs @ 17034

Last change on this file since 17034 was 17034, checked in by mkommend, 5 years ago

#2435: Updated branch with most recent trunk changes.

File size: 19.9 KB
RevLine 
[6256]1#region License Information
2/* HeuristicLab
[16565]3 * Copyright (C) 2002-2019 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
[6256]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
[8704]22using System;
[6256]23using System.Collections.Generic;
24using System.Linq;
[17034]25using HEAL.Attic;
[6256]26using HeuristicLab.Common;
27using HeuristicLab.Core;
28using HeuristicLab.Data;
29using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
[15448]30using HeuristicLab.Optimization;
[6256]31using HeuristicLab.Parameters;
32
33namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
[6555]34  [Item("Constant Optimization Evaluator", "Calculates Pearson R² of a symbolic regression solution and optimizes the constant used.")]
[16565]35  [StorableType("24B68851-036D-4446-BD6F-3823E9028FF4")]
[6256]36  public class SymbolicRegressionConstantOptimizationEvaluator : SymbolicRegressionSingleObjectiveEvaluator {
37    private const string ConstantOptimizationIterationsParameterName = "ConstantOptimizationIterations";
38    private const string ConstantOptimizationImprovementParameterName = "ConstantOptimizationImprovement";
39    private const string ConstantOptimizationProbabilityParameterName = "ConstantOptimizationProbability";
40    private const string ConstantOptimizationRowsPercentageParameterName = "ConstantOptimizationRowsPercentage";
[8823]41    private const string UpdateConstantsInTreeParameterName = "UpdateConstantsInSymbolicExpressionTree";
[13670]42    private const string UpdateVariableWeightsParameterName = "Update Variable Weights";
[6256]43
[15448]44    private const string FunctionEvaluationsResultParameterName = "Constants Optimization Function Evaluations";
45    private const string GradientEvaluationsResultParameterName = "Constants Optimization Gradient Evaluations";
[15483]46    private const string CountEvaluationsParameterName = "Count Function and Gradient Evaluations";
[15448]47
[6256]48    public IFixedValueParameter<IntValue> ConstantOptimizationIterationsParameter {
49      get { return (IFixedValueParameter<IntValue>)Parameters[ConstantOptimizationIterationsParameterName]; }
50    }
51    public IFixedValueParameter<DoubleValue> ConstantOptimizationImprovementParameter {
52      get { return (IFixedValueParameter<DoubleValue>)Parameters[ConstantOptimizationImprovementParameterName]; }
53    }
54    public IFixedValueParameter<PercentValue> ConstantOptimizationProbabilityParameter {
55      get { return (IFixedValueParameter<PercentValue>)Parameters[ConstantOptimizationProbabilityParameterName]; }
56    }
57    public IFixedValueParameter<PercentValue> ConstantOptimizationRowsPercentageParameter {
58      get { return (IFixedValueParameter<PercentValue>)Parameters[ConstantOptimizationRowsPercentageParameterName]; }
59    }
[8823]60    public IFixedValueParameter<BoolValue> UpdateConstantsInTreeParameter {
61      get { return (IFixedValueParameter<BoolValue>)Parameters[UpdateConstantsInTreeParameterName]; }
62    }
[13670]63    public IFixedValueParameter<BoolValue> UpdateVariableWeightsParameter {
64      get { return (IFixedValueParameter<BoolValue>)Parameters[UpdateVariableWeightsParameterName]; }
65    }
[6256]66
[15448]67    public IResultParameter<IntValue> FunctionEvaluationsResultParameter {
68      get { return (IResultParameter<IntValue>)Parameters[FunctionEvaluationsResultParameterName]; }
69    }
70    public IResultParameter<IntValue> GradientEvaluationsResultParameter {
71      get { return (IResultParameter<IntValue>)Parameters[GradientEvaluationsResultParameterName]; }
72    }
[15483]73    public IFixedValueParameter<BoolValue> CountEvaluationsParameter {
74      get { return (IFixedValueParameter<BoolValue>)Parameters[CountEvaluationsParameterName]; }
75    }
[13670]76
[15448]77
[6256]78    public IntValue ConstantOptimizationIterations {
79      get { return ConstantOptimizationIterationsParameter.Value; }
80    }
81    public DoubleValue ConstantOptimizationImprovement {
82      get { return ConstantOptimizationImprovementParameter.Value; }
83    }
84    public PercentValue ConstantOptimizationProbability {
85      get { return ConstantOptimizationProbabilityParameter.Value; }
86    }
87    public PercentValue ConstantOptimizationRowsPercentage {
88      get { return ConstantOptimizationRowsPercentageParameter.Value; }
89    }
[8823]90    public bool UpdateConstantsInTree {
91      get { return UpdateConstantsInTreeParameter.Value.Value; }
92      set { UpdateConstantsInTreeParameter.Value.Value = value; }
93    }
[6256]94
[13670]95    public bool UpdateVariableWeights {
96      get { return UpdateVariableWeightsParameter.Value.Value; }
97      set { UpdateVariableWeightsParameter.Value.Value = value; }
98    }
99
[15483]100    public bool CountEvaluations {
101      get { return CountEvaluationsParameter.Value.Value; }
102      set { CountEvaluationsParameter.Value.Value = value; }
103    }
104
[6256]105    public override bool Maximization {
106      get { return true; }
107    }
108
109    [StorableConstructor]
[16565]110    protected SymbolicRegressionConstantOptimizationEvaluator(StorableConstructorFlag _) : base(_) { }
[6256]111    protected SymbolicRegressionConstantOptimizationEvaluator(SymbolicRegressionConstantOptimizationEvaluator original, Cloner cloner)
112      : base(original, cloner) {
113    }
114    public SymbolicRegressionConstantOptimizationEvaluator()
115      : base() {
[17034]116      Parameters.Add(new FixedValueParameter<IntValue>(ConstantOptimizationIterationsParameterName, "Determines how many iterations should be calculated while optimizing the constant of a symbolic expression tree (0 indicates other or default stopping criterion).", new IntValue(10)));
117      Parameters.Add(new FixedValueParameter<DoubleValue>(ConstantOptimizationImprovementParameterName, "Determines the relative improvement which must be achieved in the constant optimization to continue with it (0 indicates other or default stopping criterion).", new DoubleValue(0)) { Hidden = true });
118      Parameters.Add(new FixedValueParameter<PercentValue>(ConstantOptimizationProbabilityParameterName, "Determines the probability that the constants are optimized", new PercentValue(1)));
119      Parameters.Add(new FixedValueParameter<PercentValue>(ConstantOptimizationRowsPercentageParameterName, "Determines the percentage of the rows which should be used for constant optimization", new PercentValue(1)));
[13916]120      Parameters.Add(new FixedValueParameter<BoolValue>(UpdateConstantsInTreeParameterName, "Determines if the constants in the tree should be overwritten by the optimized constants.", new BoolValue(true)) { Hidden = true });
121      Parameters.Add(new FixedValueParameter<BoolValue>(UpdateVariableWeightsParameterName, "Determines if the variable weights in the tree should be  optimized.", new BoolValue(true)) { Hidden = true });
[15448]122
[15483]123      Parameters.Add(new FixedValueParameter<BoolValue>(CountEvaluationsParameterName, "Determines if function and gradient evaluation should be counted.", new BoolValue(false)));
[15448]124      Parameters.Add(new ResultParameter<IntValue>(FunctionEvaluationsResultParameterName, "The number of function evaluations performed by the constants optimization evaluator", "Results", new IntValue()));
125      Parameters.Add(new ResultParameter<IntValue>(GradientEvaluationsResultParameterName, "The number of gradient evaluations performed by the constants optimization evaluator", "Results", new IntValue()));
[6256]126    }
127
128    public override IDeepCloneable Clone(Cloner cloner) {
129      return new SymbolicRegressionConstantOptimizationEvaluator(this, cloner);
130    }
131
[8823]132    [StorableHook(HookType.AfterDeserialization)]
133    private void AfterDeserialization() {
134      if (!Parameters.ContainsKey(UpdateConstantsInTreeParameterName))
135        Parameters.Add(new FixedValueParameter<BoolValue>(UpdateConstantsInTreeParameterName, "Determines if the constants in the tree should be overwritten by the optimized constants.", new BoolValue(true)));
[13670]136      if (!Parameters.ContainsKey(UpdateVariableWeightsParameterName))
137        Parameters.Add(new FixedValueParameter<BoolValue>(UpdateVariableWeightsParameterName, "Determines if the variable weights in the tree should be  optimized.", new BoolValue(true)));
[15448]138
[15483]139      if (!Parameters.ContainsKey(CountEvaluationsParameterName))
140        Parameters.Add(new FixedValueParameter<BoolValue>(CountEvaluationsParameterName, "Determines if function and gradient evaluation should be counted.", new BoolValue(false)));
141
[15448]142      if (!Parameters.ContainsKey(FunctionEvaluationsResultParameterName))
143        Parameters.Add(new ResultParameter<IntValue>(FunctionEvaluationsResultParameterName, "The number of function evaluations performed by the constants optimization evaluator", "Results", new IntValue()));
144      if (!Parameters.ContainsKey(GradientEvaluationsResultParameterName))
145        Parameters.Add(new ResultParameter<IntValue>(GradientEvaluationsResultParameterName, "The number of gradient evaluations performed by the constants optimization evaluator", "Results", new IntValue()));
[8823]146    }
147
[15448]148    private static readonly object locker = new object();
[10291]149    public override IOperation InstrumentedApply() {
[6256]150      var solution = SymbolicExpressionTreeParameter.ActualValue;
151      double quality;
152      if (RandomParameter.ActualValue.NextDouble() < ConstantOptimizationProbability.Value) {
153        IEnumerable<int> constantOptimizationRows = GenerateRowsToEvaluate(ConstantOptimizationRowsPercentage.Value);
[15448]154        var counter = new EvaluationsCounter();
[6256]155        quality = OptimizeConstants(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, ProblemDataParameter.ActualValue,
[15448]156           constantOptimizationRows, ApplyLinearScalingParameter.ActualValue.Value, ConstantOptimizationIterations.Value, updateVariableWeights: UpdateVariableWeights, lowerEstimationLimit: EstimationLimitsParameter.ActualValue.Lower, upperEstimationLimit: EstimationLimitsParameter.ActualValue.Upper, updateConstantsInTree: UpdateConstantsInTree, counter: counter);
[8938]157
[6256]158        if (ConstantOptimizationRowsPercentage.Value != RelativeNumberOfEvaluatedSamplesParameter.ActualValue.Value) {
159          var evaluationRows = GenerateRowsToEvaluate();
[8664]160          quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, ProblemDataParameter.ActualValue, evaluationRows, ApplyLinearScalingParameter.ActualValue.Value);
[6256]161        }
[15448]162
[15483]163        if (CountEvaluations) {
164          lock (locker) {
165            FunctionEvaluationsResultParameter.ActualValue.Value += counter.FunctionEvaluations;
166            GradientEvaluationsResultParameter.ActualValue.Value += counter.GradientEvaluations;
167          }
[15448]168        }
169
[6256]170      } else {
171        var evaluationRows = GenerateRowsToEvaluate();
[8664]172        quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, ProblemDataParameter.ActualValue, evaluationRows, ApplyLinearScalingParameter.ActualValue.Value);
[6256]173      }
174      QualityParameter.ActualValue = new DoubleValue(quality);
175
[10291]176      return base.InstrumentedApply();
[6256]177    }
178
179    public override double Evaluate(IExecutionContext context, ISymbolicExpressionTree tree, IRegressionProblemData problemData, IEnumerable<int> rows) {
180      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = context;
181      EstimationLimitsParameter.ExecutionContext = context;
[8664]182      ApplyLinearScalingParameter.ExecutionContext = context;
[15448]183      FunctionEvaluationsResultParameter.ExecutionContext = context;
184      GradientEvaluationsResultParameter.ExecutionContext = context;
[6256]185
[9209]186      // Pearson R² evaluator is used on purpose instead of the const-opt evaluator,
187      // because Evaluate() is used to get the quality of evolved models on
188      // different partitions of the dataset (e.g., best validation model)
[8664]189      double r2 = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, tree, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, problemData, rows, ApplyLinearScalingParameter.ActualValue.Value);
[6256]190
191      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = null;
192      EstimationLimitsParameter.ExecutionContext = null;
[9209]193      ApplyLinearScalingParameter.ExecutionContext = null;
[15448]194      FunctionEvaluationsResultParameter.ExecutionContext = null;
195      GradientEvaluationsResultParameter.ExecutionContext = null;
[6256]196
197      return r2;
198    }
199
[15448]200    public class EvaluationsCounter {
201      public int FunctionEvaluations = 0;
202      public int GradientEvaluations = 0;
203    }
204
[14826]205    public static double OptimizeConstants(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter,
206      ISymbolicExpressionTree tree, IRegressionProblemData problemData, IEnumerable<int> rows, bool applyLinearScaling,
207      int maxIterations, bool updateVariableWeights = true,
208      double lowerEstimationLimit = double.MinValue, double upperEstimationLimit = double.MaxValue,
[15448]209      bool updateConstantsInTree = true, Action<double[], double, object> iterationCallback = null, EvaluationsCounter counter = null) {
[8704]210
[14826]211      // numeric constants in the tree become variables for constant opt
212      // variables in the tree become parameters (fixed values) for constant opt
213      // for each parameter (variable in the original tree) we store the
214      // variable name, variable value (for factor vars) and lag as a DataForVariable object.
215      // A dictionary is used to find parameters
[14840]216      double[] initialConstants;
[14843]217      var parameters = new List<TreeToAutoDiffTermConverter.DataForVariable>();
[14826]218
[14843]219      TreeToAutoDiffTermConverter.ParametricFunction func;
220      TreeToAutoDiffTermConverter.ParametricFunctionGradient func_grad;
[15447]221      if (!TreeToAutoDiffTermConverter.TryConvertToAutoDiff(tree, updateVariableWeights, applyLinearScaling, out parameters, out initialConstants, out func, out func_grad))
[8828]222        throw new NotSupportedException("Could not optimize constants of symbolic expression tree due to not supported symbols used in the tree.");
[14826]223      if (parameters.Count == 0) return 0.0; // gkronber: constant expressions always have a R² of 0.0
224      var parameterEntries = parameters.ToArray(); // order of entries must be the same for x
[14400]225
[13670]226      //extract inital constants
[15447]227      double[] c;
228      if (applyLinearScaling) {
229        c = new double[initialConstants.Length + 2];
[15481]230        c[0] = 0.0;
231        c[1] = 1.0;
232        Array.Copy(initialConstants, 0, c, 2, initialConstants.Length);
[15447]233      } else {
234        c = (double[])initialConstants.Clone();
[14400]235      }
[15447]236
[8938]237      double originalQuality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling);
[6256]238
[15448]239      if (counter == null) counter = new EvaluationsCounter();
240      var rowEvaluationsCounter = new EvaluationsCounter();
241
[8704]242      alglib.lsfitstate state;
243      alglib.lsfitreport rep;
[14826]244      int retVal;
[6256]245
[12509]246      IDataset ds = problemData.Dataset;
[14826]247      double[,] x = new double[rows.Count(), parameters.Count];
[8704]248      int row = 0;
249      foreach (var r in rows) {
[14826]250        int col = 0;
[14840]251        foreach (var info in parameterEntries) {
[14826]252          if (ds.VariableHasType<double>(info.variableName)) {
[14946]253            x[row, col] = ds.GetDoubleValue(info.variableName, r + info.lag);
[14826]254          } else if (ds.VariableHasType<string>(info.variableName)) {
255            x[row, col] = ds.GetStringValue(info.variableName, r) == info.variableValue ? 1 : 0;
256          } else throw new InvalidProgramException("found a variable of unknown type");
257          col++;
[8704]258        }
259        row++;
260      }
261      double[] y = ds.GetDoubleValues(problemData.TargetVariable, rows).ToArray();
262      int n = x.GetLength(0);
263      int m = x.GetLength(1);
264      int k = c.Length;
[6256]265
[14840]266      alglib.ndimensional_pfunc function_cx_1_func = CreatePFunc(func);
267      alglib.ndimensional_pgrad function_cx_1_grad = CreatePGrad(func_grad);
[15371]268      alglib.ndimensional_rep xrep = (p, f, obj) => iterationCallback(p, f, obj);
[6256]269
[8704]270      try {
271        alglib.lsfitcreatefg(x, y, c, n, m, k, false, out state);
[8938]272        alglib.lsfitsetcond(state, 0.0, 0.0, maxIterations);
[15371]273        alglib.lsfitsetxrep(state, iterationCallback != null);
[8938]274        //alglib.lsfitsetgradientcheck(state, 0.001);
[15448]275        alglib.lsfitfit(state, function_cx_1_func, function_cx_1_grad, xrep, rowEvaluationsCounter);
[14826]276        alglib.lsfitresults(state, out retVal, out c, out rep);
[15447]277      } catch (ArithmeticException) {
[8984]278        return originalQuality;
[15447]279      } catch (alglib.alglibexception) {
[8984]280        return originalQuality;
[8704]281      }
[8823]282
[15448]283      counter.FunctionEvaluations += rowEvaluationsCounter.FunctionEvaluations / n;
284      counter.GradientEvaluations += rowEvaluationsCounter.GradientEvaluations / n;
285
[14826]286      //retVal == -7  => constant optimization failed due to wrong gradient
[15447]287      if (retVal != -7) {
[15481]288        if (applyLinearScaling) {
289          var tmp = new double[c.Length - 2];
290          Array.Copy(c, 2, tmp, 0, tmp.Length);
291          UpdateConstants(tree, tmp, updateVariableWeights);
292        } else UpdateConstants(tree, c, updateVariableWeights);
[15447]293      }
[8938]294      var quality = SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling);
295
[15480]296      if (!updateConstantsInTree) UpdateConstants(tree, initialConstants, updateVariableWeights);
[15447]297
[8938]298      if (originalQuality - quality > 0.001 || double.IsNaN(quality)) {
[15480]299        UpdateConstants(tree, initialConstants, updateVariableWeights);
[8938]300        return originalQuality;
[8704]301      }
[8938]302      return quality;
[6256]303    }
304
[13670]305    private static void UpdateConstants(ISymbolicExpressionTree tree, double[] constants, bool updateVariableWeights) {
[8938]306      int i = 0;
307      foreach (var node in tree.Root.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>()) {
308        ConstantTreeNode constantTreeNode = node as ConstantTreeNode;
[14951]309        VariableTreeNodeBase variableTreeNodeBase = node as VariableTreeNodeBase;
[14826]310        FactorVariableTreeNode factorVarTreeNode = node as FactorVariableTreeNode;
[8938]311        if (constantTreeNode != null)
312          constantTreeNode.Value = constants[i++];
[14951]313        else if (updateVariableWeights && variableTreeNodeBase != null)
314          variableTreeNodeBase.Weight = constants[i++];
[14826]315        else if (factorVarTreeNode != null) {
316          for (int j = 0; j < factorVarTreeNode.Weights.Length; j++)
317            factorVarTreeNode.Weights[j] = constants[i++];
318        }
[8938]319      }
320    }
321
[14843]322    private static alglib.ndimensional_pfunc CreatePFunc(TreeToAutoDiffTermConverter.ParametricFunction func) {
[14840]323      return (double[] c, double[] x, ref double fx, object o) => {
324        fx = func(c, x);
[15448]325        var counter = (EvaluationsCounter)o;
326        counter.FunctionEvaluations++;
[8704]327      };
328    }
[6256]329
[14843]330    private static alglib.ndimensional_pgrad CreatePGrad(TreeToAutoDiffTermConverter.ParametricFunctionGradient func_grad) {
[14840]331      return (double[] c, double[] x, ref double fx, double[] grad, object o) => {
[15480]332        var tuple = func_grad(c, x);
333        fx = tuple.Item2;
334        Array.Copy(tuple.Item1, grad, grad.Length);
[15448]335        var counter = (EvaluationsCounter)o;
336        counter.GradientEvaluations++;
[6256]337      };
338    }
[8730]339    public static bool CanOptimizeConstants(ISymbolicExpressionTree tree) {
[14843]340      return TreeToAutoDiffTermConverter.IsCompatible(tree);
[8730]341    }
[6256]342  }
343}
Note: See TracBrowser for help on using the repository browser.