Changeset 17006
 Timestamp:
 06/11/19 23:41:18 (5 years ago)
 File:

 1 edited
Legend:
 Unmodified
 Added
 Removed

branches/2994AutoDiffForIntervals/HeuristicLab.Problems.DataAnalysis.Regression.Symbolic.Extensions/ConstrainedConstantOptimizationEvaluator.cs
r17004 r17006 74 74 get { return (IFixedValueParameter<BoolValue>)Parameters[CountEvaluationsParameterName]; } 75 75 } 76 public IConstrainedValueParameter<StringValue> SolverParameter { 77 get { return (IConstrainedValueParameter<StringValue>)Parameters["Solver"]; } 78 } 76 79 77 80 … … 103 106 } 104 107 108 public string Solver { 109 get { return SolverParameter.Value.Value; } 110 } 105 111 public override bool Maximization { 106 112 get { return false; } … … 122 128 123 129 Parameters.Add(new FixedValueParameter<BoolValue>(CountEvaluationsParameterName, "Determines if function and gradient evaluation should be counted.", new BoolValue(false))); 130 var validSolvers = new ItemSet<StringValue>(new[] { "nonsmooth (minns)", "sequential linear programming (minnlc)" }.Select(s => new StringValue(s).AsReadOnly())); 131 Parameters.Add(new ConstrainedValueParameter<StringValue>("Solver", "The solver algorithm", validSolvers, validSolvers.First())); 124 132 Parameters.Add(new ResultParameter<IntValue>(FunctionEvaluationsResultParameterName, "The number of function evaluations performed by the constants optimization evaluator", "Results", new IntValue())); 125 133 Parameters.Add(new ResultParameter<IntValue>(GradientEvaluationsResultParameterName, "The number of gradient evaluations performed by the constants optimization evaluator", "Results", new IntValue())); … … 142 150 var counter = new EvaluationsCounter(); 143 151 quality = OptimizeConstants(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, ProblemDataParameter.ActualValue, 144 constantOptimizationRows, ApplyLinearScalingParameter.ActualValue.Value, ConstantOptimizationIterations.Value, updateVariableWeights: UpdateVariableWeights, lowerEstimationLimit: EstimationLimitsParameter.ActualValue.Lower, upperEstimationLimit: EstimationLimitsParameter.ActualValue.Upper, updateConstantsInTree: UpdateConstantsInTree, counter: counter);152 constantOptimizationRows, ApplyLinearScalingParameter.ActualValue.Value, Solver, ConstantOptimizationIterations.Value, updateVariableWeights: UpdateVariableWeights, lowerEstimationLimit: EstimationLimitsParameter.ActualValue.Lower, upperEstimationLimit: EstimationLimitsParameter.ActualValue.Upper, updateConstantsInTree: UpdateConstantsInTree, counter: counter); 145 153 146 154 if (ConstantOptimizationRowsPercentage.Value != RelativeNumberOfEvaluatedSamplesParameter.ActualValue.Value) { … … 210 218 public static double OptimizeConstants(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, 211 219 ISymbolicExpressionTree tree, IRegressionProblemData problemData, IEnumerable<int> rows, bool applyLinearScaling, 220 string solver, 212 221 int maxIterations, bool updateVariableWeights = true, 213 222 double lowerEstimationLimit = double.MinValue, double upperEstimationLimit = double.MaxValue, … … 323 332 } 324 333 325 326 327 alglib.minnsstate state;328 alglib.minnsreport rep;329 try {330 alglib.minnscreate(thetaValues.Count, thetaValues.ToArray(), out state);331 // alglib.minnssetalgoslp(state); // SLP is more robust but slower332 alglib.minnssetbc(state, thetaValues.Select(_ => 10000.0).ToArray(), thetaValues.Select(_ => +10000.0).ToArray());333 alglib.minnssetcond(state, 1E7, maxIterations);334 var s = Enumerable.Repeat(1d, thetaValues.Count).ToArray(); // scale is set to unit scale335 alglib.minnssetscale(state, s); 336 337 // set nonlinear constraints: 0 equality constraints, constraintTrees inequality constraints338 alglib.minnssetnlc(state, 0, constraintTrees.Count); 339 340 alglib.minnsoptimize(state, calculate_jacobian, null, null);341 alglib.minnsresults(state, out double[] xOpt, out rep); 342 343 344 // counter.FunctionEvaluations += rep.nfev; TODO345 counter.GradientEvaluations += rep.nfev; 346 347 if (rep.terminationtype != 8) {348 // update parameters in tree349 var pIdx = 0;350 foreach (var node in tree.IterateNodesPostfix()) {351 if(node is ConstantTreeNode constTreeNode) {352 constTreeNode.Value = xOpt[pIdx++];353 } else if(node is VariableTreeNode varTreeNode) {354 varTreeNode.Weight = xOpt[pIdx++];334 if (solver.Contains("minns")) { 335 alglib.minnsstate state; 336 alglib.minnsreport rep; 337 try { 338 alglib.minnscreate(thetaValues.Count, thetaValues.ToArray(), out state); 339 // alglib.minnssetalgoslp(state); // SLP is more robust but slower 340 alglib.minnssetbc(state, thetaValues.Select(_ => 10000.0).ToArray(), thetaValues.Select(_ => +10000.0).ToArray()); 341 alglib.minnssetcond(state, 1E7, maxIterations); 342 var s = Enumerable.Repeat(1d, thetaValues.Count).ToArray(); // scale is set to unit scale 343 alglib.minnssetscale(state, s); 344 345 // set nonlinear constraints: 0 equality constraints, constraintTrees inequality constraints 346 alglib.minnssetnlc(state, 0, constraintTrees.Count); 347 348 alglib.minnsoptimize(state, calculate_jacobian, null, null); 349 alglib.minnsresults(state, out double[] xOpt, out rep); 350 351 352 // counter.FunctionEvaluations += rep.nfev; TODO 353 counter.GradientEvaluations += rep.nfev; 354 355 if (rep.terminationtype != 8) { 356 // update parameters in tree 357 var pIdx = 0; 358 foreach (var node in tree.IterateNodesPostfix()) { 359 if (node is ConstantTreeNode constTreeNode) { 360 constTreeNode.Value = xOpt[pIdx++]; 361 } else if (node is VariableTreeNode varTreeNode) { 362 varTreeNode.Weight = xOpt[pIdx++]; 363 } 355 364 } 356 } 357 358 // note: we keep the optimized constants even when the tree is worse. 359 } 360 361 } catch (ArithmeticException) { 362 // eval MSE of original tree 363 return SymbolicRegressionSingleObjectiveMeanSquaredErrorEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling: false); 364 365 } catch (alglib.alglibexception) { 366 // eval MSE of original tree 367 return SymbolicRegressionSingleObjectiveMeanSquaredErrorEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling: false); 368 } 369 365 366 // note: we keep the optimized constants even when the tree is worse. 367 } 368 369 } catch (ArithmeticException) { 370 // eval MSE of original tree 371 return SymbolicRegressionSingleObjectiveMeanSquaredErrorEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling: false); 372 373 } catch (alglib.alglibexception) { 374 // eval MSE of original tree 375 return SymbolicRegressionSingleObjectiveMeanSquaredErrorEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling: false); 376 } 377 } else if (solver.Contains("minnlc")) { 378 alglib.minnlcstate state; 379 alglib.minnlcreport rep; 380 alglib.optguardreport optGuardRep; 381 try { 382 alglib.minnlccreate(thetaValues.Count, thetaValues.ToArray(), out state); 383 alglib.minnlcsetalgoslp(state); // SLP is more robust but slower 384 alglib.minnlcsetbc(state, thetaValues.Select(_ => 10000.0).ToArray(), thetaValues.Select(_ => +10000.0).ToArray()); 385 alglib.minnlcsetcond(state, 1E7, maxIterations); 386 var s = Enumerable.Repeat(1d, thetaValues.Count).ToArray(); // scale is set to unit scale 387 alglib.minnlcsetscale(state, s); 388 389 // set nonlinear constraints: 0 equality constraints, constraintTrees inequality constraints 390 alglib.minnlcsetnlc(state, 0, constraintTrees.Count); 391 alglib.minnlcoptguardsmoothness(state, 1); 392 393 alglib.minnlcoptimize(state, calculate_jacobian, null, null); 394 alglib.minnlcresults(state, out double[] xOpt, out rep); 395 alglib.minnlcoptguardresults(state, out optGuardRep); 396 if (optGuardRep.nonc0suspected) throw new InvalidProgramException("optGuardRep.nonc0suspected"); 397 if (optGuardRep.nonc1suspected) throw new InvalidProgramException("optGuardRep.nonc1suspected"); 398 399 // counter.FunctionEvaluations += rep.nfev; TODO 400 counter.GradientEvaluations += rep.nfev; 401 402 if (rep.terminationtype != 8) { 403 // update parameters in tree 404 var pIdx = 0; 405 foreach (var node in tree.IterateNodesPostfix()) { 406 if (node is ConstantTreeNode constTreeNode) { 407 constTreeNode.Value = xOpt[pIdx++]; 408 } else if (node is VariableTreeNode varTreeNode) { 409 varTreeNode.Weight = xOpt[pIdx++]; 410 } 411 } 412 413 // note: we keep the optimized constants even when the tree is worse. 414 } 415 416 } catch (ArithmeticException) { 417 // eval MSE of original tree 418 return SymbolicRegressionSingleObjectiveMeanSquaredErrorEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling: false); 419 420 } catch (alglib.alglibexception) { 421 // eval MSE of original tree 422 return SymbolicRegressionSingleObjectiveMeanSquaredErrorEvaluator.Calculate(interpreter, tree, lowerEstimationLimit, upperEstimationLimit, problemData, rows, applyLinearScaling: false); 423 } 424 } else { 425 throw new ArgumentException($"Unknown solver {solver}"); 426 } 427 370 428 371 429 // evaluate tree with updated constants
Note: See TracChangeset
for help on using the changeset viewer.