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 |
|
---|
22 | using System;
|
---|
23 | using System.Collections.Generic;
|
---|
24 | using System.Linq;
|
---|
25 | using System.Threading;
|
---|
26 | using HeuristicLab.Common;
|
---|
27 | using HeuristicLab.Core;
|
---|
28 | using HeuristicLab.Data;
|
---|
29 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
30 | using HeuristicLab.Parameters;
|
---|
31 | using HEAL.Attic;
|
---|
32 | using DoubleVector = MathNet.Numerics.LinearAlgebra.Vector<double>;
|
---|
33 |
|
---|
34 | namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
|
---|
35 | [StorableType("5F8DB251-C6F7-40AC-BC30-3C55AF655A2F")]
|
---|
36 | [Item("VectorUnrollingNonlinearLeastSquaresConstantOptimizationEvaluator", "")]
|
---|
37 | public class VectorUnrollingNonlinearLeastSquaresConstantOptimizationEvaluator : SymbolicRegressionConstantOptimizationEvaluator {
|
---|
38 |
|
---|
39 | private const string ConstantOptimizationIterationsName = "ConstantOptimizationIterations";
|
---|
40 |
|
---|
41 | #region Parameter Properties
|
---|
42 | public IFixedValueParameter<IntValue> ConstantOptimizationIterationsParameter {
|
---|
43 | get { return (IFixedValueParameter<IntValue>)Parameters[ConstantOptimizationIterationsName]; }
|
---|
44 | }
|
---|
45 | #endregion
|
---|
46 |
|
---|
47 | #region Properties
|
---|
48 | public int ConstantOptimizationIterations {
|
---|
49 | get { return ConstantOptimizationIterationsParameter.Value.Value; }
|
---|
50 | set { ConstantOptimizationIterationsParameter.Value.Value = value; }
|
---|
51 | }
|
---|
52 | #endregion
|
---|
53 |
|
---|
54 | public VectorUnrollingNonlinearLeastSquaresConstantOptimizationEvaluator()
|
---|
55 | : base() {
|
---|
56 | Parameters.Add(new FixedValueParameter<IntValue>(ConstantOptimizationIterationsName, "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)));
|
---|
57 | }
|
---|
58 |
|
---|
59 | protected VectorUnrollingNonlinearLeastSquaresConstantOptimizationEvaluator(VectorUnrollingNonlinearLeastSquaresConstantOptimizationEvaluator original, Cloner cloner)
|
---|
60 | : base(original, cloner) {
|
---|
61 | }
|
---|
62 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
63 | return new VectorUnrollingNonlinearLeastSquaresConstantOptimizationEvaluator(this, cloner);
|
---|
64 | }
|
---|
65 | [StorableConstructor]
|
---|
66 | protected VectorUnrollingNonlinearLeastSquaresConstantOptimizationEvaluator(StorableConstructorFlag _) : base(_) { }
|
---|
67 |
|
---|
68 | protected override ISymbolicExpressionTree OptimizeConstants(
|
---|
69 | ISymbolicExpressionTree tree, IRegressionProblemData problemData, IEnumerable<int> rows,
|
---|
70 | CancellationToken cancellationToken = default(CancellationToken), EvaluationsCounter counter = null) {
|
---|
71 | return OptimizeTree(tree, (SymbolicDataAnalysisExpressionTreeVectorInterpreter)SymbolicDataAnalysisTreeInterpreterParameter.ActualValue,
|
---|
72 | problemData, rows,
|
---|
73 | ApplyLinearScalingParameter.ActualValue.Value, ConstantOptimizationIterations, UpdateVariableWeights,
|
---|
74 | cancellationToken, counter);
|
---|
75 | }
|
---|
76 |
|
---|
77 | public static ISymbolicExpressionTree OptimizeTree(
|
---|
78 | ISymbolicExpressionTree tree,
|
---|
79 | SymbolicDataAnalysisExpressionTreeVectorInterpreter interpreter,
|
---|
80 | IRegressionProblemData problemData, IEnumerable<int> rows,
|
---|
81 | bool applyLinearScaling, int maxIterations, bool updateVariableWeights,
|
---|
82 | CancellationToken cancellationToken = default(CancellationToken), EvaluationsCounter counter = null, Action<double[], double, object> iterationCallback = null) {
|
---|
83 |
|
---|
84 | var vectorLengths = problemData.Dataset.DoubleVectorVariables
|
---|
85 | .SelectMany(var => problemData.Dataset.GetDoubleVectorValues(var, rows))
|
---|
86 | .Select(v => v.Count);
|
---|
87 | if (vectorLengths.Distinct().Count() > 1)
|
---|
88 | throw new InvalidOperationException("All vectors must be of same length.");
|
---|
89 | var evaluationTraces = interpreter.GetIntermediateNodeValues(tree, problemData.Dataset, rows);
|
---|
90 | var evaluationTrace = evaluationTraces.First(); // assume all vector lengths are the same
|
---|
91 |
|
---|
92 |
|
---|
93 | // numeric constants in the tree become variables for constant opt
|
---|
94 | // variables in the tree become parameters (fixed values) for constant opt
|
---|
95 | // for each parameter (variable in the original tree) we store the
|
---|
96 | // variable name, variable value (for factor vars) and lag as a DataForVariable object.
|
---|
97 | // A dictionary is used to find parameters
|
---|
98 | bool success = VectorUnrollingTreeToAutoDiffTermConverter.TryConvertToAutoDiff(
|
---|
99 | tree, evaluationTrace,
|
---|
100 | updateVariableWeights, applyLinearScaling,
|
---|
101 | out var parameters, out var initialConstants, out var func, out var func_grad);
|
---|
102 | if (!success)
|
---|
103 | throw new NotSupportedException("Could not optimize constants of symbolic expression tree due to not supported symbols used in the tree.");
|
---|
104 | if (parameters.Count == 0) return (ISymbolicExpressionTree)tree.Clone();
|
---|
105 | var parameterEntries = parameters.ToArray(); // order of entries must be the same for x
|
---|
106 |
|
---|
107 | //extract initial constants
|
---|
108 | double[] c;
|
---|
109 | if (applyLinearScaling) {
|
---|
110 | c = new double[initialConstants.Length + 2];
|
---|
111 | c[0] = 0.0;
|
---|
112 | c[1] = 1.0;
|
---|
113 | Array.Copy(initialConstants, 0, c, 2, initialConstants.Length);
|
---|
114 | } else {
|
---|
115 | c = (double[])initialConstants.Clone();
|
---|
116 | }
|
---|
117 |
|
---|
118 | IDataset ds = problemData.Dataset;
|
---|
119 | double[,] x = new double[rows.Count(), parameters.Count];
|
---|
120 | int row = 0;
|
---|
121 | foreach (var r in rows) {
|
---|
122 | int col = 0;
|
---|
123 | foreach (var info in parameterEntries) {
|
---|
124 | if (ds.VariableHasType<double>(info.variableName)) {
|
---|
125 | x[row, col] = ds.GetDoubleValue(info.variableName, r + info.lag);
|
---|
126 | } else if (ds.VariableHasType<string>(info.variableName)) {
|
---|
127 | x[row, col] = ds.GetStringValue(info.variableName, r) == info.variableValue ? 1 : 0;
|
---|
128 | } else if (ds.VariableHasType<DoubleVector>(info.variableName)) {
|
---|
129 | x[row, col] = ds.GetDoubleVectorValue(info.variableName, r)[info.index];
|
---|
130 | } else throw new InvalidProgramException("found a variable of unknown type");
|
---|
131 | col++;
|
---|
132 | }
|
---|
133 | row++;
|
---|
134 | }
|
---|
135 | double[] y = ds.GetDoubleValues(problemData.TargetVariable, rows).ToArray();
|
---|
136 | int n = x.GetLength(0);
|
---|
137 | int m = x.GetLength(1);
|
---|
138 | int k = c.Length;
|
---|
139 |
|
---|
140 | alglib.ndimensional_pfunc function_cx_1_func = CreatePFunc(func);
|
---|
141 | alglib.ndimensional_pgrad function_cx_1_grad = CreatePGrad(func_grad);
|
---|
142 | alglib.ndimensional_rep xrep = (p, f, obj) => {
|
---|
143 | iterationCallback?.Invoke(p, f, obj);
|
---|
144 | cancellationToken.ThrowIfCancellationRequested();
|
---|
145 | };
|
---|
146 | var rowEvaluationsCounter = new EvaluationsCounter();
|
---|
147 |
|
---|
148 | try {
|
---|
149 | alglib.lsfitcreatefg(x, y, c, n, m, k, false, out var state);
|
---|
150 | alglib.lsfitsetcond(state, 0.0, 0.0, maxIterations);
|
---|
151 | alglib.lsfitsetxrep(state, iterationCallback != null || cancellationToken != default(CancellationToken));
|
---|
152 | //alglib.lsfitsetgradientcheck(state, 0.001);
|
---|
153 | alglib.lsfitfit(state, function_cx_1_func, function_cx_1_grad, xrep, rowEvaluationsCounter);
|
---|
154 | alglib.lsfitresults(state, out var retVal, out c, out alglib.lsfitreport rep);
|
---|
155 |
|
---|
156 | //retVal == -7 => constant optimization failed due to wrong gradient
|
---|
157 | if (retVal == -1)
|
---|
158 | return (ISymbolicExpressionTree)tree.Clone();
|
---|
159 | } catch (ArithmeticException) {
|
---|
160 | return (ISymbolicExpressionTree)tree.Clone();
|
---|
161 | } catch (alglib.alglibexception) {
|
---|
162 | return (ISymbolicExpressionTree)tree.Clone();
|
---|
163 | }
|
---|
164 |
|
---|
165 | if (counter != null) {
|
---|
166 | counter.FunctionEvaluations += rowEvaluationsCounter.FunctionEvaluations / n;
|
---|
167 | counter.GradientEvaluations += rowEvaluationsCounter.GradientEvaluations / n;
|
---|
168 | }
|
---|
169 |
|
---|
170 | var newTree = (ISymbolicExpressionTree)tree.Clone();
|
---|
171 | if (applyLinearScaling) {
|
---|
172 | var tmp = new double[c.Length - 2];
|
---|
173 | Array.Copy(c, 2, tmp, 0, tmp.Length);
|
---|
174 | UpdateConstants(newTree, tmp, updateVariableWeights);
|
---|
175 | } else
|
---|
176 | UpdateConstants(newTree, c, updateVariableWeights);
|
---|
177 |
|
---|
178 | return newTree;
|
---|
179 | }
|
---|
180 |
|
---|
181 | private static void UpdateConstants(ISymbolicExpressionTree tree, double[] constants, bool updateVariableWeights) {
|
---|
182 | int i = 0;
|
---|
183 | foreach (var node in tree.Root.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>()) {
|
---|
184 | ConstantTreeNode constantTreeNode = node as ConstantTreeNode;
|
---|
185 | VariableTreeNodeBase variableTreeNodeBase = node as VariableTreeNodeBase;
|
---|
186 | FactorVariableTreeNode factorVarTreeNode = node as FactorVariableTreeNode;
|
---|
187 | if (constantTreeNode != null)
|
---|
188 | constantTreeNode.Value = constants[i++];
|
---|
189 | else if (updateVariableWeights && variableTreeNodeBase != null)
|
---|
190 | variableTreeNodeBase.Weight = constants[i++];
|
---|
191 | else if (factorVarTreeNode != null) {
|
---|
192 | for (int j = 0; j < factorVarTreeNode.Weights.Length; j++)
|
---|
193 | factorVarTreeNode.Weights[j] = constants[i++];
|
---|
194 | }
|
---|
195 | }
|
---|
196 | }
|
---|
197 |
|
---|
198 | private static alglib.ndimensional_pfunc CreatePFunc(VectorUnrollingTreeToAutoDiffTermConverter.ParametricFunction func) {
|
---|
199 | return (double[] c, double[] x, ref double fx, object o) => {
|
---|
200 | fx = func(c, x);
|
---|
201 | var counter = (EvaluationsCounter)o;
|
---|
202 | counter.FunctionEvaluations++;
|
---|
203 | };
|
---|
204 | }
|
---|
205 |
|
---|
206 | private static alglib.ndimensional_pgrad CreatePGrad(VectorUnrollingTreeToAutoDiffTermConverter.ParametricFunctionGradient func_grad) {
|
---|
207 | return (double[] c, double[] x, ref double fx, double[] grad, object o) => {
|
---|
208 | var tuple = func_grad(c, x);
|
---|
209 | fx = tuple.Item2;
|
---|
210 | Array.Copy(tuple.Item1, grad, grad.Length);
|
---|
211 | var counter = (EvaluationsCounter)o;
|
---|
212 | counter.GradientEvaluations++;
|
---|
213 | };
|
---|
214 | }
|
---|
215 |
|
---|
216 | public static bool CanOptimizeConstants(ISymbolicExpressionTree tree) {
|
---|
217 | return VectorUnrollingTreeToAutoDiffTermConverter.IsCompatible(tree);
|
---|
218 | }
|
---|
219 | }
|
---|
220 | } |
---|