[13645] | 1 | #region License Information
|
---|
| 2 | /* HeuristicLab
|
---|
[14185] | 3 | * Copyright (C) 2002-2016 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
|
---|
[13645] | 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.Diagnostics.Contracts;
|
---|
| 25 | using System.Linq;
|
---|
[13658] | 26 | using HeuristicLab.Algorithms.DataAnalysis.MctsSymbolicRegression.Policies;
|
---|
[13645] | 27 | using HeuristicLab.Core;
|
---|
| 28 | using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
|
---|
| 29 | using HeuristicLab.Problems.DataAnalysis;
|
---|
| 30 | using HeuristicLab.Problems.DataAnalysis.Symbolic;
|
---|
| 31 | using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
|
---|
| 32 | using HeuristicLab.Random;
|
---|
[14929] | 33 | using HeuristicLab.Persistence;
|
---|
[13645] | 34 |
|
---|
| 35 | namespace HeuristicLab.Algorithms.DataAnalysis.MctsSymbolicRegression {
|
---|
| 36 | public static class MctsSymbolicRegressionStatic {
|
---|
| 37 | // TODO: SGD with adagrad instead of lbfgs?
|
---|
| 38 | // TODO: check Taylor expansion capabilities (ln(x), sqrt(x), exp(x)) in combination with GBT
|
---|
| 39 | // TODO: optimize for 3 targets concurrently (y, 1/y, exp(y), and log(y))? Would simplify the number of possible expressions again
|
---|
| 40 | #region static API
|
---|
| 41 |
|
---|
[14929] | 42 | [StorableType("c370f13f-cfd9-4388-be24-09a979d74740")]
|
---|
[13645] | 43 | public interface IState {
|
---|
| 44 | bool Done { get; }
|
---|
| 45 | ISymbolicRegressionModel BestModel { get; }
|
---|
| 46 | double BestSolutionTrainingQuality { get; }
|
---|
| 47 | double BestSolutionTestQuality { get; }
|
---|
[13651] | 48 | int TotalRollouts { get; }
|
---|
| 49 | int EffectiveRollouts { get; }
|
---|
| 50 | int FuncEvaluations { get; }
|
---|
| 51 | int GradEvaluations { get; } // number of gradient evaluations (* num parameters) to get a value representative of the effort comparable to the number of function evaluations
|
---|
| 52 | // TODO other stats on LM optimizer might be interesting here
|
---|
[13645] | 53 | }
|
---|
| 54 |
|
---|
| 55 | // created through factory method
|
---|
| 56 | private class State : IState {
|
---|
| 57 | private const int MaxParams = 100;
|
---|
| 58 |
|
---|
| 59 | // state variables used by MCTS
|
---|
| 60 | internal readonly Automaton automaton;
|
---|
| 61 | internal IRandom random { get; private set; }
|
---|
| 62 | internal readonly Tree tree;
|
---|
| 63 | internal readonly Func<byte[], int, double> evalFun;
|
---|
[13658] | 64 | internal readonly IPolicy treePolicy;
|
---|
[13651] | 65 | // MCTS might get stuck. Track statistics on the number of effective rollouts
|
---|
| 66 | internal int totalRollouts;
|
---|
| 67 | internal int effectiveRollouts;
|
---|
[13645] | 68 |
|
---|
| 69 |
|
---|
| 70 | // state variables used only internally (for eval function)
|
---|
| 71 | private readonly IRegressionProblemData problemData;
|
---|
| 72 | private readonly double[][] x;
|
---|
| 73 | private readonly double[] y;
|
---|
| 74 | private readonly double[][] testX;
|
---|
| 75 | private readonly double[] testY;
|
---|
| 76 | private readonly double[] scalingFactor;
|
---|
| 77 | private readonly double[] scalingOffset;
|
---|
| 78 | private readonly int constOptIterations;
|
---|
| 79 | private readonly double lowerEstimationLimit, upperEstimationLimit;
|
---|
| 80 |
|
---|
| 81 | private readonly ExpressionEvaluator evaluator, testEvaluator;
|
---|
| 82 |
|
---|
| 83 | // values for best solution
|
---|
| 84 | private double bestRSq;
|
---|
| 85 | private byte[] bestCode;
|
---|
| 86 | private int bestNParams;
|
---|
| 87 | private double[] bestConsts;
|
---|
| 88 |
|
---|
[13651] | 89 | // stats
|
---|
| 90 | private int funcEvaluations;
|
---|
| 91 | private int gradEvaluations;
|
---|
| 92 |
|
---|
[13645] | 93 | // buffers
|
---|
| 94 | private readonly double[] ones; // vector of ones (as default params)
|
---|
| 95 | private readonly double[] constsBuf;
|
---|
| 96 | private readonly double[] predBuf, testPredBuf;
|
---|
| 97 | private readonly double[][] gradBuf;
|
---|
| 98 |
|
---|
[13658] | 99 | public State(IRegressionProblemData problemData, uint randSeed, int maxVariables, bool scaleVariables, int constOptIterations,
|
---|
| 100 | IPolicy treePolicy = null,
|
---|
[13645] | 101 | double lowerEstimationLimit = double.MinValue, double upperEstimationLimit = double.MaxValue,
|
---|
| 102 | bool allowProdOfVars = true,
|
---|
| 103 | bool allowExp = true,
|
---|
| 104 | bool allowLog = true,
|
---|
| 105 | bool allowInv = true,
|
---|
| 106 | bool allowMultipleTerms = false) {
|
---|
| 107 |
|
---|
| 108 | this.problemData = problemData;
|
---|
| 109 | this.constOptIterations = constOptIterations;
|
---|
| 110 | this.evalFun = this.Eval;
|
---|
| 111 | this.lowerEstimationLimit = lowerEstimationLimit;
|
---|
| 112 | this.upperEstimationLimit = upperEstimationLimit;
|
---|
| 113 |
|
---|
| 114 | random = new MersenneTwister(randSeed);
|
---|
| 115 |
|
---|
| 116 | // prepare data for evaluation
|
---|
| 117 | double[][] x;
|
---|
| 118 | double[] y;
|
---|
| 119 | double[][] testX;
|
---|
| 120 | double[] testY;
|
---|
| 121 | double[] scalingFactor;
|
---|
| 122 | double[] scalingOffset;
|
---|
| 123 | // get training and test datasets (scale linearly based on training set if required)
|
---|
| 124 | GenerateData(problemData, scaleVariables, problemData.TrainingIndices, out x, out y, out scalingFactor, out scalingOffset);
|
---|
| 125 | GenerateData(problemData, problemData.TestIndices, scalingFactor, scalingOffset, out testX, out testY);
|
---|
| 126 | this.x = x;
|
---|
| 127 | this.y = y;
|
---|
| 128 | this.testX = testX;
|
---|
| 129 | this.testY = testY;
|
---|
| 130 | this.scalingFactor = scalingFactor;
|
---|
| 131 | this.scalingOffset = scalingOffset;
|
---|
| 132 | this.evaluator = new ExpressionEvaluator(y.Length, lowerEstimationLimit, upperEstimationLimit);
|
---|
| 133 | // we need a separate evaluator because the vector length for the test dataset might differ
|
---|
| 134 | this.testEvaluator = new ExpressionEvaluator(testY.Length, lowerEstimationLimit, upperEstimationLimit);
|
---|
| 135 |
|
---|
| 136 | this.automaton = new Automaton(x, maxVariables, allowProdOfVars, allowExp, allowLog, allowInv, allowMultipleTerms);
|
---|
[13658] | 137 | this.treePolicy = treePolicy ?? new Ucb();
|
---|
| 138 | this.tree = new Tree() { state = automaton.CurrentState, actionStatistics = treePolicy.CreateActionStatistics() };
|
---|
[13645] | 139 |
|
---|
| 140 | // reset best solution
|
---|
| 141 | this.bestRSq = 0;
|
---|
| 142 | // code for default solution (constant model)
|
---|
| 143 | this.bestCode = new byte[] { (byte)OpCodes.LoadConst0, (byte)OpCodes.Exit };
|
---|
| 144 | this.bestNParams = 0;
|
---|
| 145 | this.bestConsts = null;
|
---|
| 146 |
|
---|
| 147 | // init buffers
|
---|
| 148 | this.ones = Enumerable.Repeat(1.0, MaxParams).ToArray();
|
---|
| 149 | constsBuf = new double[MaxParams];
|
---|
| 150 | this.predBuf = new double[y.Length];
|
---|
| 151 | this.testPredBuf = new double[testY.Length];
|
---|
| 152 |
|
---|
| 153 | this.gradBuf = Enumerable.Range(0, MaxParams).Select(_ => new double[y.Length]).ToArray();
|
---|
| 154 | }
|
---|
| 155 |
|
---|
| 156 | #region IState inferface
|
---|
[13658] | 157 | public bool Done { get { return tree != null && tree.Done; } }
|
---|
[13645] | 158 |
|
---|
| 159 | public double BestSolutionTrainingQuality {
|
---|
| 160 | get {
|
---|
| 161 | evaluator.Exec(bestCode, x, bestConsts, predBuf);
|
---|
| 162 | return RSq(y, predBuf);
|
---|
| 163 | }
|
---|
| 164 | }
|
---|
| 165 |
|
---|
| 166 | public double BestSolutionTestQuality {
|
---|
| 167 | get {
|
---|
| 168 | testEvaluator.Exec(bestCode, testX, bestConsts, testPredBuf);
|
---|
| 169 | return RSq(testY, testPredBuf);
|
---|
| 170 | }
|
---|
| 171 | }
|
---|
| 172 |
|
---|
| 173 | // takes the code of the best solution and creates and equivalent symbolic regression model
|
---|
| 174 | public ISymbolicRegressionModel BestModel {
|
---|
| 175 | get {
|
---|
| 176 | var treeGen = new SymbolicExpressionTreeGenerator(problemData.AllowedInputVariables.ToArray());
|
---|
| 177 | var interpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
|
---|
| 178 |
|
---|
| 179 | var t = new SymbolicExpressionTree(treeGen.Exec(bestCode, bestConsts, bestNParams, scalingFactor, scalingOffset));
|
---|
[13941] | 180 | var model = new SymbolicRegressionModel(problemData.TargetVariable, t, interpreter, lowerEstimationLimit, upperEstimationLimit);
|
---|
[13645] | 181 |
|
---|
| 182 | // model has already been scaled linearly in Eval
|
---|
| 183 | return model;
|
---|
| 184 | }
|
---|
| 185 | }
|
---|
[13651] | 186 |
|
---|
| 187 | public int TotalRollouts { get { return totalRollouts; } }
|
---|
| 188 | public int EffectiveRollouts { get { return effectiveRollouts; } }
|
---|
| 189 | public int FuncEvaluations { get { return funcEvaluations; } }
|
---|
| 190 | public int GradEvaluations { get { return gradEvaluations; } } // number of gradient evaluations (* num parameters) to get a value representative of the effort comparable to the number of function evaluations
|
---|
| 191 |
|
---|
[13645] | 192 | #endregion
|
---|
| 193 |
|
---|
| 194 | private double Eval(byte[] code, int nParams) {
|
---|
| 195 | double[] optConsts;
|
---|
| 196 | double q;
|
---|
| 197 | Eval(code, nParams, out q, out optConsts);
|
---|
| 198 |
|
---|
| 199 | if (q > bestRSq) {
|
---|
| 200 | bestRSq = q;
|
---|
| 201 | bestNParams = nParams;
|
---|
| 202 | this.bestCode = new byte[code.Length];
|
---|
| 203 | this.bestConsts = new double[bestNParams];
|
---|
| 204 |
|
---|
| 205 | Array.Copy(code, bestCode, code.Length);
|
---|
| 206 | Array.Copy(optConsts, bestConsts, bestNParams);
|
---|
| 207 | }
|
---|
| 208 |
|
---|
| 209 | return q;
|
---|
| 210 | }
|
---|
| 211 |
|
---|
| 212 | private void Eval(byte[] code, int nParams, out double rsq, out double[] optConsts) {
|
---|
| 213 | // we make a first pass to determine a valid starting configuration for all constants
|
---|
| 214 | // constant c in log(c + f(x)) is adjusted to guarantee that x is positive (see expression evaluator)
|
---|
| 215 | // scale and offset are set to optimal starting configuration
|
---|
| 216 | // assumes scale is the first param and offset is the last param
|
---|
| 217 | double alpha;
|
---|
| 218 | double beta;
|
---|
| 219 |
|
---|
| 220 | // reset constants
|
---|
| 221 | Array.Copy(ones, constsBuf, nParams);
|
---|
| 222 | evaluator.Exec(code, x, constsBuf, predBuf, adjustOffsetForLogAndExp: true);
|
---|
[13651] | 223 | funcEvaluations++;
|
---|
[13645] | 224 |
|
---|
| 225 | // calc opt scaling (alpha*f(x) + beta)
|
---|
| 226 | OnlineCalculatorError error;
|
---|
| 227 | OnlineLinearScalingParameterCalculator.Calculate(predBuf, y, out alpha, out beta, out error);
|
---|
| 228 | if (error == OnlineCalculatorError.None) {
|
---|
| 229 | constsBuf[0] *= beta;
|
---|
| 230 | constsBuf[nParams - 1] = constsBuf[nParams - 1] * beta + alpha;
|
---|
| 231 | }
|
---|
| 232 | if (nParams <= 2 || constOptIterations <= 0) {
|
---|
| 233 | // if we don't need to optimize parameters then we are done
|
---|
| 234 | // changing scale and offset does not influence r²
|
---|
| 235 | rsq = RSq(y, predBuf);
|
---|
| 236 | optConsts = constsBuf;
|
---|
| 237 | } else {
|
---|
| 238 | // optimize constants using the starting point calculated above
|
---|
| 239 | OptimizeConstsLm(code, constsBuf, nParams, 0.0, nIters: constOptIterations);
|
---|
[13651] | 240 |
|
---|
[13645] | 241 | evaluator.Exec(code, x, constsBuf, predBuf);
|
---|
[13651] | 242 | funcEvaluations++;
|
---|
| 243 |
|
---|
[13645] | 244 | rsq = RSq(y, predBuf);
|
---|
| 245 | optConsts = constsBuf;
|
---|
| 246 | }
|
---|
| 247 | }
|
---|
| 248 |
|
---|
| 249 |
|
---|
| 250 |
|
---|
| 251 | #region helpers
|
---|
| 252 | private static double RSq(IEnumerable<double> x, IEnumerable<double> y) {
|
---|
| 253 | OnlineCalculatorError error;
|
---|
| 254 | double r = OnlinePearsonsRCalculator.Calculate(x, y, out error);
|
---|
| 255 | return error == OnlineCalculatorError.None ? r * r : 0.0;
|
---|
| 256 | }
|
---|
| 257 |
|
---|
| 258 |
|
---|
| 259 | private void OptimizeConstsLm(byte[] code, double[] consts, int nParams, double epsF = 0.0, int nIters = 100) {
|
---|
[13651] | 260 | double[] optConsts = new double[nParams]; // allocate a smaller buffer for constants opt (TODO perf?)
|
---|
[13645] | 261 | Array.Copy(consts, optConsts, nParams);
|
---|
| 262 |
|
---|
| 263 | alglib.minlmstate state;
|
---|
| 264 | alglib.minlmreport rep = null;
|
---|
[13657] | 265 | alglib.minlmcreatevj(y.Length, optConsts, out state);
|
---|
[13645] | 266 | alglib.minlmsetcond(state, 0.0, epsF, 0.0, nIters);
|
---|
| 267 | //alglib.minlmsetgradientcheck(state, 0.000001);
|
---|
| 268 | alglib.minlmoptimize(state, Func, FuncAndJacobian, null, code);
|
---|
| 269 | alglib.minlmresults(state, out optConsts, out rep);
|
---|
[13651] | 270 | funcEvaluations += rep.nfunc;
|
---|
| 271 | gradEvaluations += rep.njac * nParams;
|
---|
[13645] | 272 |
|
---|
| 273 | if (rep.terminationtype < 0) throw new ArgumentException("lm failed: termination type = " + rep.terminationtype);
|
---|
| 274 |
|
---|
| 275 | // only use optimized constants if successful
|
---|
| 276 | if (rep.terminationtype >= 0) {
|
---|
| 277 | Array.Copy(optConsts, consts, optConsts.Length);
|
---|
| 278 | }
|
---|
| 279 | }
|
---|
| 280 |
|
---|
| 281 | private void Func(double[] arg, double[] fi, object obj) {
|
---|
| 282 | var code = (byte[])obj;
|
---|
| 283 | evaluator.Exec(code, x, arg, predBuf); // gradients are nParams x vLen
|
---|
| 284 | for (int r = 0; r < predBuf.Length; r++) {
|
---|
| 285 | var res = predBuf[r] - y[r];
|
---|
| 286 | fi[r] = res;
|
---|
| 287 | }
|
---|
| 288 | }
|
---|
| 289 | private void FuncAndJacobian(double[] arg, double[] fi, double[,] jac, object obj) {
|
---|
| 290 | int nParams = arg.Length;
|
---|
| 291 | var code = (byte[])obj;
|
---|
| 292 | evaluator.ExecGradient(code, x, arg, predBuf, gradBuf); // gradients are nParams x vLen
|
---|
| 293 | for (int r = 0; r < predBuf.Length; r++) {
|
---|
| 294 | var res = predBuf[r] - y[r];
|
---|
| 295 | fi[r] = res;
|
---|
| 296 |
|
---|
| 297 | for (int k = 0; k < nParams; k++) {
|
---|
| 298 | jac[r, k] = gradBuf[k][r];
|
---|
| 299 | }
|
---|
| 300 | }
|
---|
| 301 | }
|
---|
| 302 | #endregion
|
---|
| 303 | }
|
---|
| 304 |
|
---|
[13658] | 305 | public static IState CreateState(IRegressionProblemData problemData, uint randSeed, int maxVariables = 3,
|
---|
| 306 | bool scaleVariables = true, int constOptIterations = 0,
|
---|
| 307 | IPolicy policy = null,
|
---|
| 308 | double lowerEstimationLimit = double.MinValue, double upperEstimationLimit = double.MaxValue,
|
---|
[13645] | 309 | bool allowProdOfVars = true,
|
---|
| 310 | bool allowExp = true,
|
---|
| 311 | bool allowLog = true,
|
---|
| 312 | bool allowInv = true,
|
---|
| 313 | bool allowMultipleTerms = false
|
---|
| 314 | ) {
|
---|
[13658] | 315 | return new State(problemData, randSeed, maxVariables, scaleVariables, constOptIterations,
|
---|
| 316 | policy,
|
---|
[13645] | 317 | lowerEstimationLimit, upperEstimationLimit,
|
---|
| 318 | allowProdOfVars, allowExp, allowLog, allowInv, allowMultipleTerms);
|
---|
| 319 | }
|
---|
| 320 |
|
---|
| 321 | // returns the quality of the evaluated solution
|
---|
| 322 | public static double MakeStep(IState state) {
|
---|
| 323 | var mctsState = state as State;
|
---|
| 324 | if (mctsState == null) throw new ArgumentException("state");
|
---|
| 325 | if (mctsState.Done) throw new NotSupportedException("The tree search has enumerated all possible solutions.");
|
---|
| 326 |
|
---|
| 327 | return TreeSearch(mctsState);
|
---|
| 328 | }
|
---|
| 329 | #endregion
|
---|
| 330 |
|
---|
| 331 | private static double TreeSearch(State mctsState) {
|
---|
| 332 | var automaton = mctsState.automaton;
|
---|
| 333 | var tree = mctsState.tree;
|
---|
| 334 | var eval = mctsState.evalFun;
|
---|
| 335 | var rand = mctsState.random;
|
---|
[13658] | 336 | var treePolicy = mctsState.treePolicy;
|
---|
[13651] | 337 | double q = 0;
|
---|
| 338 | bool success = false;
|
---|
| 339 | do {
|
---|
| 340 | automaton.Reset();
|
---|
[13658] | 341 | success = TryTreeSearchRec(rand, tree, automaton, eval, treePolicy, out q);
|
---|
[13651] | 342 | mctsState.totalRollouts++;
|
---|
[13658] | 343 | } while (!success && !tree.Done);
|
---|
[13651] | 344 | mctsState.effectiveRollouts++;
|
---|
| 345 | return q;
|
---|
[13645] | 346 | }
|
---|
| 347 |
|
---|
[13651] | 348 | // tree search might fail because of constraints for expressions
|
---|
| 349 | // in this case we get stuck we just restart
|
---|
| 350 | // see ConstraintHandler.cs for more info
|
---|
[13658] | 351 | private static bool TryTreeSearchRec(IRandom rand, Tree tree, Automaton automaton, Func<byte[], int, double> eval, IPolicy treePolicy,
|
---|
| 352 | out double q) {
|
---|
[13645] | 353 | Tree selectedChild = null;
|
---|
| 354 | Contract.Assert(tree.state == automaton.CurrentState);
|
---|
[13658] | 355 | Contract.Assert(!tree.Done);
|
---|
[13645] | 356 | if (tree.children == null) {
|
---|
| 357 | if (automaton.IsFinalState(tree.state)) {
|
---|
| 358 | // final state
|
---|
[13658] | 359 | tree.Done = true;
|
---|
[13645] | 360 |
|
---|
| 361 | // EVALUATE
|
---|
| 362 | byte[] code; int nParams;
|
---|
| 363 | automaton.GetCode(out code, out nParams);
|
---|
| 364 | q = eval(code, nParams);
|
---|
[13658] | 365 |
|
---|
| 366 | treePolicy.Update(tree.actionStatistics, q);
|
---|
[13651] | 367 | return true; // we reached a final state
|
---|
[13645] | 368 | } else {
|
---|
| 369 | // EXPAND
|
---|
| 370 | int[] possibleFollowStates;
|
---|
| 371 | int nFs;
|
---|
| 372 | automaton.FollowStates(automaton.CurrentState, out possibleFollowStates, out nFs);
|
---|
[13651] | 373 | if (nFs == 0) {
|
---|
| 374 | // stuck in a dead end (no final state and no allowed follow states)
|
---|
| 375 | q = 0;
|
---|
[13658] | 376 | tree.Done = true;
|
---|
[13651] | 377 | tree.children = null;
|
---|
| 378 | return false;
|
---|
| 379 | }
|
---|
[13645] | 380 | tree.children = new Tree[nFs];
|
---|
[13651] | 381 | for (int i = 0; i < tree.children.Length; i++)
|
---|
[13658] | 382 | tree.children[i] = new Tree() { children = null, state = possibleFollowStates[i], actionStatistics = treePolicy.CreateActionStatistics() };
|
---|
[13645] | 383 |
|
---|
[13657] | 384 | selectedChild = nFs > 1 ? SelectFinalOrRandom(automaton, tree, rand) : tree.children[0];
|
---|
[13645] | 385 | }
|
---|
| 386 | } else {
|
---|
| 387 | // tree.children != null
|
---|
| 388 | // UCT selection within tree
|
---|
[13658] | 389 | int selectedIdx = 0;
|
---|
| 390 | if (tree.children.Length > 1) {
|
---|
| 391 | selectedIdx = treePolicy.Select(tree.children.Select(ch => ch.actionStatistics), rand);
|
---|
| 392 | }
|
---|
| 393 | selectedChild = tree.children[selectedIdx];
|
---|
[13645] | 394 | }
|
---|
| 395 | // make selected step and recurse
|
---|
| 396 | automaton.Goto(selectedChild.state);
|
---|
[13658] | 397 | var success = TryTreeSearchRec(rand, selectedChild, automaton, eval, treePolicy, out q);
|
---|
[13651] | 398 | if (success) {
|
---|
| 399 | // only update if successful
|
---|
[13658] | 400 | treePolicy.Update(tree.actionStatistics, q);
|
---|
[13651] | 401 | }
|
---|
[13645] | 402 |
|
---|
[13658] | 403 | tree.Done = tree.children.All(ch => ch.Done);
|
---|
| 404 | if (tree.Done) {
|
---|
[13651] | 405 | tree.children = null; // cut off the sub-branch if it has been fully explored
|
---|
[13645] | 406 | }
|
---|
[13651] | 407 | return success;
|
---|
[13645] | 408 | }
|
---|
| 409 |
|
---|
| 410 | private static Tree SelectFinalOrRandom(Automaton automaton, Tree tree, IRandom rand) {
|
---|
| 411 | // if one of the new children leads to a final state then go there
|
---|
| 412 | // otherwise choose a random child
|
---|
| 413 | int selectedChildIdx = -1;
|
---|
| 414 | // find first final state if there is one
|
---|
| 415 | for (int i = 0; i < tree.children.Length; i++) {
|
---|
| 416 | if (automaton.IsFinalState(tree.children[i].state)) {
|
---|
| 417 | selectedChildIdx = i;
|
---|
| 418 | break;
|
---|
| 419 | }
|
---|
| 420 | }
|
---|
[13669] | 421 | // no final state -> select a the first child
|
---|
[13645] | 422 | if (selectedChildIdx == -1) {
|
---|
[13669] | 423 | selectedChildIdx = 0;
|
---|
[13645] | 424 | }
|
---|
| 425 | return tree.children[selectedChildIdx];
|
---|
| 426 | }
|
---|
| 427 |
|
---|
| 428 | // scales data and extracts values from dataset into arrays
|
---|
| 429 | private static void GenerateData(IRegressionProblemData problemData, bool scaleVariables, IEnumerable<int> rows,
|
---|
| 430 | out double[][] xs, out double[] y, out double[] scalingFactor, out double[] scalingOffset) {
|
---|
| 431 | xs = new double[problemData.AllowedInputVariables.Count()][];
|
---|
| 432 |
|
---|
| 433 | var i = 0;
|
---|
| 434 | if (scaleVariables) {
|
---|
| 435 | scalingFactor = new double[xs.Length];
|
---|
| 436 | scalingOffset = new double[xs.Length];
|
---|
| 437 | } else {
|
---|
| 438 | scalingFactor = null;
|
---|
| 439 | scalingOffset = null;
|
---|
| 440 | }
|
---|
| 441 | foreach (var var in problemData.AllowedInputVariables) {
|
---|
| 442 | if (scaleVariables) {
|
---|
| 443 | var minX = problemData.Dataset.GetDoubleValues(var, rows).Min();
|
---|
| 444 | var maxX = problemData.Dataset.GetDoubleValues(var, rows).Max();
|
---|
| 445 | var range = maxX - minX;
|
---|
| 446 |
|
---|
| 447 | // scaledX = (x - min) / range
|
---|
| 448 | var sf = 1.0 / range;
|
---|
| 449 | var offset = -minX / range;
|
---|
| 450 | scalingFactor[i] = sf;
|
---|
| 451 | scalingOffset[i] = offset;
|
---|
| 452 | i++;
|
---|
| 453 | }
|
---|
| 454 | }
|
---|
| 455 |
|
---|
| 456 | GenerateData(problemData, rows, scalingFactor, scalingOffset, out xs, out y);
|
---|
| 457 | }
|
---|
| 458 |
|
---|
| 459 | // extract values from dataset into arrays
|
---|
| 460 | private static void GenerateData(IRegressionProblemData problemData, IEnumerable<int> rows, double[] scalingFactor, double[] scalingOffset,
|
---|
| 461 | out double[][] xs, out double[] y) {
|
---|
| 462 | xs = new double[problemData.AllowedInputVariables.Count()][];
|
---|
| 463 |
|
---|
| 464 | int i = 0;
|
---|
| 465 | foreach (var var in problemData.AllowedInputVariables) {
|
---|
| 466 | var sf = scalingFactor == null ? 1.0 : scalingFactor[i];
|
---|
| 467 | var offset = scalingFactor == null ? 0.0 : scalingOffset[i];
|
---|
| 468 | xs[i++] =
|
---|
| 469 | problemData.Dataset.GetDoubleValues(var, rows).Select(xi => xi * sf + offset).ToArray();
|
---|
| 470 | }
|
---|
| 471 |
|
---|
| 472 | y = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows).ToArray();
|
---|
| 473 | }
|
---|
| 474 | }
|
---|
| 475 | }
|
---|