1 | #region License Information
|
---|
2 | /* HeuristicLab
|
---|
3 | * Copyright (C) 2002-2015 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.Linq;
|
---|
24 | using System.Runtime.CompilerServices;
|
---|
25 | using System.Threading;
|
---|
26 | using HeuristicLab.Analysis;
|
---|
27 | using HeuristicLab.Common;
|
---|
28 | using HeuristicLab.Core;
|
---|
29 | using HeuristicLab.Data;
|
---|
30 | using HeuristicLab.Optimization;
|
---|
31 | using HeuristicLab.Parameters;
|
---|
32 | using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
|
---|
33 | using HeuristicLab.Problems.DataAnalysis;
|
---|
34 |
|
---|
35 | namespace HeuristicLab.Algorithms.DataAnalysis.MctsSymbolicRegression {
|
---|
36 | [Item("MCTS Symbolic Regression", "Monte carlo tree search for symbolic regression. Useful mainly as a base learner in gradient boosting.")]
|
---|
37 | [StorableClass]
|
---|
38 | [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 250)]
|
---|
39 | public class MctsSymbolicRegressionAlgorithm : BasicAlgorithm {
|
---|
40 | public override Type ProblemType {
|
---|
41 | get { return typeof(IRegressionProblem); }
|
---|
42 | }
|
---|
43 | public new IRegressionProblem Problem {
|
---|
44 | get { return (IRegressionProblem)base.Problem; }
|
---|
45 | set { base.Problem = value; }
|
---|
46 | }
|
---|
47 |
|
---|
48 | #region ParameterNames
|
---|
49 | private const string IterationsParameterName = "Iterations";
|
---|
50 | private const string MaxVariablesParameterName = "Maximum variables";
|
---|
51 | private const string ScaleVariablesParameterName = "Scale variables";
|
---|
52 | private const string AllowedFactorsParameterName = "Allowed factors";
|
---|
53 | private const string ConstantOptimizationIterationsParameterName = "Iterations (constant optimization)";
|
---|
54 | private const string CParameterName = "C";
|
---|
55 | private const string SeedParameterName = "Seed";
|
---|
56 | private const string SetSeedRandomlyParameterName = "SetSeedRandomly";
|
---|
57 | private const string UpdateIntervalParameterName = "UpdateInterval";
|
---|
58 | private const string CreateSolutionParameterName = "CreateSolution";
|
---|
59 | private const string PunishmentFactorParameterName = "PunishmentFactor";
|
---|
60 |
|
---|
61 | private const string VariableProductFactorName = "product(xi)";
|
---|
62 | private const string ExpFactorName = "exp(c * product(xi))";
|
---|
63 | private const string LogFactorName = "log(c + sum(c*product(xi))";
|
---|
64 | private const string InvFactorName = "1 / (1 + sum(c*product(xi))";
|
---|
65 | private const string FactorSumsName = "sum of multiple terms";
|
---|
66 | #endregion
|
---|
67 |
|
---|
68 | #region ParameterProperties
|
---|
69 | public IFixedValueParameter<IntValue> IterationsParameter {
|
---|
70 | get { return (IFixedValueParameter<IntValue>)Parameters[IterationsParameterName]; }
|
---|
71 | }
|
---|
72 | public IFixedValueParameter<IntValue> MaxSizeParameter {
|
---|
73 | get { return (IFixedValueParameter<IntValue>)Parameters[MaxVariablesParameterName]; }
|
---|
74 | }
|
---|
75 | public IFixedValueParameter<BoolValue> ScaleVariablesParameter {
|
---|
76 | get { return (IFixedValueParameter<BoolValue>)Parameters[ScaleVariablesParameterName]; }
|
---|
77 | }
|
---|
78 | public IFixedValueParameter<IntValue> ConstantOptimizationIterationsParameter {
|
---|
79 | get { return (IFixedValueParameter<IntValue>)Parameters[ConstantOptimizationIterationsParameterName]; }
|
---|
80 | }
|
---|
81 | public IFixedValueParameter<DoubleValue> CParameter {
|
---|
82 | get { return (IFixedValueParameter<DoubleValue>)Parameters[CParameterName]; }
|
---|
83 | }
|
---|
84 | public IFixedValueParameter<DoubleValue> PunishmentFactorParameter {
|
---|
85 | get { return (IFixedValueParameter<DoubleValue>)Parameters[PunishmentFactorParameterName]; }
|
---|
86 | }
|
---|
87 | public IValueParameter<ICheckedItemList<StringValue>> AllowedFactorsParameter {
|
---|
88 | get { return (IValueParameter<ICheckedItemList<StringValue>>)Parameters[AllowedFactorsParameterName]; }
|
---|
89 | }
|
---|
90 | public IFixedValueParameter<IntValue> SeedParameter {
|
---|
91 | get { return (IFixedValueParameter<IntValue>)Parameters[SeedParameterName]; }
|
---|
92 | }
|
---|
93 | public FixedValueParameter<BoolValue> SetSeedRandomlyParameter {
|
---|
94 | get { return (FixedValueParameter<BoolValue>)Parameters[SetSeedRandomlyParameterName]; }
|
---|
95 | }
|
---|
96 | public IFixedValueParameter<IntValue> UpdateIntervalParameter {
|
---|
97 | get { return (IFixedValueParameter<IntValue>)Parameters[UpdateIntervalParameterName]; }
|
---|
98 | }
|
---|
99 | public IFixedValueParameter<BoolValue> CreateSolutionParameter {
|
---|
100 | get { return (IFixedValueParameter<BoolValue>)Parameters[CreateSolutionParameterName]; }
|
---|
101 | }
|
---|
102 | #endregion
|
---|
103 |
|
---|
104 | #region Properties
|
---|
105 | public int Iterations {
|
---|
106 | get { return IterationsParameter.Value.Value; }
|
---|
107 | set { IterationsParameter.Value.Value = value; }
|
---|
108 | }
|
---|
109 | public int Seed {
|
---|
110 | get { return SeedParameter.Value.Value; }
|
---|
111 | set { SeedParameter.Value.Value = value; }
|
---|
112 | }
|
---|
113 | public bool SetSeedRandomly {
|
---|
114 | get { return SetSeedRandomlyParameter.Value.Value; }
|
---|
115 | set { SetSeedRandomlyParameter.Value.Value = value; }
|
---|
116 | }
|
---|
117 | public int MaxSize {
|
---|
118 | get { return MaxSizeParameter.Value.Value; }
|
---|
119 | set { MaxSizeParameter.Value.Value = value; }
|
---|
120 | }
|
---|
121 | public double C {
|
---|
122 | get { return CParameter.Value.Value; }
|
---|
123 | set { CParameter.Value.Value = value; }
|
---|
124 | }
|
---|
125 |
|
---|
126 | public double PunishmentFactor {
|
---|
127 | get { return PunishmentFactorParameter.Value.Value; }
|
---|
128 | set { PunishmentFactorParameter.Value.Value = value; }
|
---|
129 | }
|
---|
130 | public ICheckedItemList<StringValue> AllowedFactors {
|
---|
131 | get { return AllowedFactorsParameter.Value; }
|
---|
132 | }
|
---|
133 | public int ConstantOptimizationIterations {
|
---|
134 | get { return ConstantOptimizationIterationsParameter.Value.Value; }
|
---|
135 | set { ConstantOptimizationIterationsParameter.Value.Value = value; }
|
---|
136 | }
|
---|
137 | public bool ScaleVariables {
|
---|
138 | get { return ScaleVariablesParameter.Value.Value; }
|
---|
139 | set { ScaleVariablesParameter.Value.Value = value; }
|
---|
140 | }
|
---|
141 | public bool CreateSolution {
|
---|
142 | get { return CreateSolutionParameter.Value.Value; }
|
---|
143 | set { CreateSolutionParameter.Value.Value = value; }
|
---|
144 | }
|
---|
145 | #endregion
|
---|
146 |
|
---|
147 | [StorableConstructor]
|
---|
148 | protected MctsSymbolicRegressionAlgorithm(bool deserializing) : base(deserializing) { }
|
---|
149 |
|
---|
150 | protected MctsSymbolicRegressionAlgorithm(MctsSymbolicRegressionAlgorithm original, Cloner cloner)
|
---|
151 | : base(original, cloner) {
|
---|
152 | }
|
---|
153 |
|
---|
154 | public override IDeepCloneable Clone(Cloner cloner) {
|
---|
155 | return new MctsSymbolicRegressionAlgorithm(this, cloner);
|
---|
156 | }
|
---|
157 |
|
---|
158 | public MctsSymbolicRegressionAlgorithm() {
|
---|
159 | Problem = new RegressionProblem(); // default problem
|
---|
160 |
|
---|
161 | var defaultFactorsList = new CheckedItemList<StringValue>(
|
---|
162 | new string[] { VariableProductFactorName, ExpFactorName, LogFactorName, InvFactorName, FactorSumsName }
|
---|
163 | .Select(s => new StringValue(s).AsReadOnly())
|
---|
164 | ).AsReadOnly();
|
---|
165 | defaultFactorsList.SetItemCheckedState(defaultFactorsList.First(s => s.Value == FactorSumsName), false);
|
---|
166 |
|
---|
167 | Parameters.Add(new FixedValueParameter<IntValue>(IterationsParameterName,
|
---|
168 | "Number of iterations", new IntValue(100000)));
|
---|
169 | Parameters.Add(new FixedValueParameter<IntValue>(SeedParameterName,
|
---|
170 | "The random seed used to initialize the new pseudo random number generator.", new IntValue(0)));
|
---|
171 | Parameters.Add(new FixedValueParameter<BoolValue>(SetSeedRandomlyParameterName,
|
---|
172 | "True if the random seed should be set to a random value, otherwise false.", new BoolValue(true)));
|
---|
173 | Parameters.Add(new FixedValueParameter<IntValue>(MaxVariablesParameterName,
|
---|
174 | "Maximal number of variables references in the symbolic regression models (multiple usages of the same variable are counted)", new IntValue(5)));
|
---|
175 | Parameters.Add(new FixedValueParameter<DoubleValue>(CParameterName,
|
---|
176 | "Balancing parameter in UCT formula (0 < c < 1000). Small values: greedy search. Large values: enumeration. Default: 1.0", new DoubleValue(1.0)));
|
---|
177 | Parameters.Add(new ValueParameter<ICheckedItemList<StringValue>>(AllowedFactorsParameterName,
|
---|
178 | "Choose which expressions are allowed as factors in the model.", defaultFactorsList));
|
---|
179 |
|
---|
180 | Parameters.Add(new FixedValueParameter<IntValue>(ConstantOptimizationIterationsParameterName,
|
---|
181 | "Number of iterations for constant optimization. A small number of iterations should be sufficient for most models. " +
|
---|
182 | "Set to 0 to disable constants optimization.", new IntValue(10)));
|
---|
183 | Parameters.Add(new FixedValueParameter<BoolValue>(ScaleVariablesParameterName,
|
---|
184 | "Set to true to scale all input variables to the range [0..1]", new BoolValue(false)));
|
---|
185 | Parameters[ScaleVariablesParameterName].Hidden = true;
|
---|
186 | Parameters.Add(new FixedValueParameter<DoubleValue>(PunishmentFactorParameterName, "Estimations of models can be bounded. The estimation limits are calculated in the following way (lb = mean(y) - punishmentFactor*range(y), ub = mean(y) + punishmentFactor*range(y))", new DoubleValue(10)));
|
---|
187 | Parameters[PunishmentFactorParameterName].Hidden = true;
|
---|
188 | Parameters.Add(new FixedValueParameter<IntValue>(UpdateIntervalParameterName,
|
---|
189 | "Number of iterations until the results are updated", new IntValue(100)));
|
---|
190 | Parameters[UpdateIntervalParameterName].Hidden = true;
|
---|
191 | Parameters.Add(new FixedValueParameter<BoolValue>(CreateSolutionParameterName,
|
---|
192 | "Flag that indicates if a solution should be produced at the end of the run", new BoolValue(true)));
|
---|
193 | Parameters[CreateSolutionParameterName].Hidden = true;
|
---|
194 | }
|
---|
195 |
|
---|
196 | [StorableHook(HookType.AfterDeserialization)]
|
---|
197 | private void AfterDeserialization() {
|
---|
198 | }
|
---|
199 |
|
---|
200 | protected override void Run(CancellationToken cancellationToken) {
|
---|
201 | // Set up the algorithm
|
---|
202 | if (SetSeedRandomly) Seed = new System.Random().Next();
|
---|
203 |
|
---|
204 | // Set up the results display
|
---|
205 | var iterations = new IntValue(0);
|
---|
206 | Results.Add(new Result("Iterations", iterations));
|
---|
207 |
|
---|
208 | var table = new DataTable("Qualities");
|
---|
209 | table.Rows.Add(new DataRow("Best quality"));
|
---|
210 | table.Rows.Add(new DataRow("Current best quality"));
|
---|
211 | table.Rows.Add(new DataRow("Average quality"));
|
---|
212 | Results.Add(new Result("Qualities", table));
|
---|
213 |
|
---|
214 | var bestQuality = new DoubleValue();
|
---|
215 | Results.Add(new Result("Best quality", bestQuality));
|
---|
216 |
|
---|
217 | var curQuality = new DoubleValue();
|
---|
218 | Results.Add(new Result("Current best quality", curQuality));
|
---|
219 |
|
---|
220 | var avgQuality = new DoubleValue();
|
---|
221 | Results.Add(new Result("Average quality", avgQuality));
|
---|
222 |
|
---|
223 | var totalRollouts = new IntValue();
|
---|
224 | Results.Add(new Result("Total rollouts", totalRollouts));
|
---|
225 | var effRollouts = new IntValue();
|
---|
226 | Results.Add(new Result("Effective rollouts", effRollouts));
|
---|
227 | var funcEvals = new IntValue();
|
---|
228 | Results.Add(new Result("Function evaluations", funcEvals));
|
---|
229 | var gradEvals = new IntValue();
|
---|
230 | Results.Add(new Result("Gradient evaluations", gradEvals));
|
---|
231 |
|
---|
232 |
|
---|
233 | // same as in SymbolicRegressionSingleObjectiveProblem
|
---|
234 | var y = Problem.ProblemData.Dataset.GetDoubleValues(Problem.ProblemData.TargetVariable,
|
---|
235 | Problem.ProblemData.TrainingIndices);
|
---|
236 | var avgY = y.Average();
|
---|
237 | var minY = y.Min();
|
---|
238 | var maxY = y.Max();
|
---|
239 | var range = maxY - minY;
|
---|
240 | var lowerLimit = avgY - PunishmentFactor * range;
|
---|
241 | var upperLimit = avgY + PunishmentFactor * range;
|
---|
242 |
|
---|
243 | // init
|
---|
244 | var problemData = (IRegressionProblemData)Problem.ProblemData.Clone();
|
---|
245 | if (!AllowedFactors.CheckedItems.Any()) throw new ArgumentException("At least on type of factor must be allowed");
|
---|
246 | var state = MctsSymbolicRegressionStatic.CreateState(problemData, (uint)Seed, MaxSize, C, ScaleVariables, ConstantOptimizationIterations,
|
---|
247 | lowerLimit, upperLimit,
|
---|
248 | allowProdOfVars: AllowedFactors.CheckedItems.Any(s => s.Value.Value == VariableProductFactorName),
|
---|
249 | allowExp: AllowedFactors.CheckedItems.Any(s => s.Value.Value == ExpFactorName),
|
---|
250 | allowLog: AllowedFactors.CheckedItems.Any(s => s.Value.Value == LogFactorName),
|
---|
251 | allowInv: AllowedFactors.CheckedItems.Any(s => s.Value.Value == InvFactorName),
|
---|
252 | allowMultipleTerms: AllowedFactors.CheckedItems.Any(s => s.Value.Value == FactorSumsName)
|
---|
253 | );
|
---|
254 |
|
---|
255 | var updateInterval = UpdateIntervalParameter.Value.Value;
|
---|
256 | double sumQ = 0.0;
|
---|
257 | double bestQ = 0.0;
|
---|
258 | double curBestQ = 0.0;
|
---|
259 | double q = 0.0;
|
---|
260 | int n = 0;
|
---|
261 | // Loop until iteration limit reached or canceled.
|
---|
262 | for (int i = 0; i < Iterations && !state.Done; i++) {
|
---|
263 | cancellationToken.ThrowIfCancellationRequested();
|
---|
264 |
|
---|
265 | q = MctsSymbolicRegressionStatic.MakeStep(state);
|
---|
266 | sumQ += q; // sum of qs in the last updateinterval iterations
|
---|
267 | curBestQ = Math.Max(q, curBestQ); // the best q in the last updateinterval iterations
|
---|
268 | bestQ = Math.Max(q, bestQ); // the best q overall
|
---|
269 | n++;
|
---|
270 | // iteration results
|
---|
271 | if (n == updateInterval) {
|
---|
272 | bestQuality.Value = bestQ;
|
---|
273 | curQuality.Value = curBestQ;
|
---|
274 | avgQuality.Value = sumQ / n;
|
---|
275 | sumQ = 0.0;
|
---|
276 | curBestQ = 0.0;
|
---|
277 |
|
---|
278 | funcEvals.Value = state.FuncEvaluations;
|
---|
279 | gradEvals.Value = state.GradEvaluations;
|
---|
280 | effRollouts.Value = state.EffectiveRollouts;
|
---|
281 | totalRollouts.Value = state.TotalRollouts;
|
---|
282 |
|
---|
283 | table.Rows["Best quality"].Values.Add(bestQuality.Value);
|
---|
284 | table.Rows["Current best quality"].Values.Add(curQuality.Value);
|
---|
285 | table.Rows["Average quality"].Values.Add(avgQuality.Value);
|
---|
286 | iterations.Value += n;
|
---|
287 | n = 0;
|
---|
288 | }
|
---|
289 | }
|
---|
290 |
|
---|
291 | // final results
|
---|
292 | if (n > 0) {
|
---|
293 | bestQuality.Value = bestQ;
|
---|
294 | curQuality.Value = curBestQ;
|
---|
295 | avgQuality.Value = sumQ / n;
|
---|
296 |
|
---|
297 | funcEvals.Value = state.FuncEvaluations;
|
---|
298 | gradEvals.Value = state.GradEvaluations;
|
---|
299 | effRollouts.Value = state.EffectiveRollouts;
|
---|
300 | totalRollouts.Value = state.TotalRollouts;
|
---|
301 |
|
---|
302 | table.Rows["Best quality"].Values.Add(bestQuality.Value);
|
---|
303 | table.Rows["Current best quality"].Values.Add(curQuality.Value);
|
---|
304 | table.Rows["Average quality"].Values.Add(avgQuality.Value);
|
---|
305 | iterations.Value = iterations.Value + n;
|
---|
306 |
|
---|
307 | }
|
---|
308 |
|
---|
309 |
|
---|
310 | Results.Add(new Result("Best solution quality (train)", new DoubleValue(state.BestSolutionTrainingQuality)));
|
---|
311 | Results.Add(new Result("Best solution quality (test)", new DoubleValue(state.BestSolutionTestQuality)));
|
---|
312 |
|
---|
313 |
|
---|
314 | // produce solution
|
---|
315 | if (CreateSolution) {
|
---|
316 | var model = state.BestModel;
|
---|
317 |
|
---|
318 | // otherwise we produce a regression solution
|
---|
319 | Results.Add(new Result("Solution", model.CreateRegressionSolution(problemData)));
|
---|
320 | }
|
---|
321 | }
|
---|
322 | }
|
---|
323 | }
|
---|