protected SymbolicDataAnalysisProblem(T problemData, U evaluator, V solutionCreator)
: base(evaluator, solutionCreator)
{
Parameters.Add(new ValueParameter <T>(ProblemDataParameterName, ProblemDataParameterDescription, problemData));
Parameters.Add(new ValueParameter <ISymbolicDataAnalysisGrammar>(SymbolicExpressionTreeGrammarParameterName, SymbolicExpressionTreeGrammarParameterDescription));
Parameters.Add(new ValueParameter <ISymbolicDataAnalysisExpressionTreeInterpreter>(SymbolicExpressionTreeInterpreterParameterName, SymoblicExpressionTreeInterpreterParameterDescription));
Parameters.Add(new FixedValueParameter <IntValue>(MaximumSymbolicExpressionTreeDepthParameterName, MaximumSymbolicExpressionTreeDepthParameterDescription));
Parameters.Add(new FixedValueParameter <IntValue>(MaximumSymbolicExpressionTreeLengthParameterName, MaximumSymbolicExpressionTreeLengthParameterDescription));
Parameters.Add(new FixedValueParameter <IntValue>(MaximumFunctionDefinitionsParameterName, MaximumFunctionDefinitionsParameterDescription));
Parameters.Add(new FixedValueParameter <IntValue>(MaximumFunctionArgumentsParameterName, MaximumFunctionArgumentsParameterDescription));
Parameters.Add(new FixedValueParameter <IntRange>(FitnessCalculationPartitionParameterName, FitnessCalculationPartitionParameterDescription));
Parameters.Add(new FixedValueParameter <IntRange>(ValidationPartitionParameterName, ValidationPartitionParameterDescription));
Parameters.Add(new FixedValueParameter <PercentValue>(RelativeNumberOfEvaluatedSamplesParameterName, RelativeNumberOfEvaluatedSamplesParameterDescription, new PercentValue(1)));
Parameters.Add(new FixedValueParameter <BoolValue>(ApplyLinearScalingParameterName, ApplyLinearScalingParameterDescription, new BoolValue(false)));
SymbolicExpressionTreeInterpreterParameter.Hidden = true;
MaximumFunctionArgumentsParameter.Hidden = true;
MaximumFunctionDefinitionsParameter.Hidden = true;
ApplyLinearScalingParameter.Hidden = true;
SymbolicExpressionTreeGrammar = new TypeCoherentExpressionGrammar();
SymbolicExpressionTreeInterpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
FitnessCalculationPartition.Start = ProblemData.TrainingPartition.Start;
FitnessCalculationPartition.End = ProblemData.TrainingPartition.End;
InitializeOperators();
UpdateGrammar();
RegisterEventHandlers();
}
public static ISymbolicExpressionGrammar CreateArithmeticAndAdfGrammar() {
var g = new TypeCoherentExpressionGrammar();
g.ConfigureAsDefaultRegressionGrammar();
g.Symbols.OfType<Variable>().First().Enabled = false;
g.MaximumFunctionArguments = 3;
g.MinimumFunctionArguments = 0;
g.MaximumFunctionDefinitions = 3;
g.MinimumFunctionDefinitions = 0;
return g;
}
public void MaxCommonSubtreeSimilarityCalculatorTestPerformance() {
var grammar = new TypeCoherentExpressionGrammar();
grammar.ConfigureAsDefaultRegressionGrammar();
var twister = new MersenneTwister(31415);
var ds = Util.CreateRandomDataset(twister, Rows, Columns);
var trees = Util.CreateRandomTrees(twister, ds, grammar, N, 1, 100, 0, 0);
double s = 0;
var sw = new Stopwatch();
sw.Start();
for (int i = 0; i < trees.Length - 1; ++i) {
for (int j = i + 1; j < trees.Length; ++j) {
s += SymbolicExpressionTreeMaxCommonSubtreeSimilarityCalculator.MaxCommonSubtreeSimilarity(trees[i], trees[j], comparer);
}
}
sw.Stop();
Console.WriteLine("Elapsed time: " + sw.ElapsedMilliseconds / 1000.0 + ", Avg. similarity: " + s / (N * (N - 1) / 2));
Console.WriteLine(N * (N + 1) / (2 * sw.ElapsedMilliseconds / 1000.0) + " similarity calculations per second.");
}
private GeneticAlgorithm CreateGpSymbolicClassificationSample() {
GeneticAlgorithm ga = new GeneticAlgorithm();
#region Problem Configuration
SymbolicClassificationSingleObjectiveProblem symbClassProblem = new SymbolicClassificationSingleObjectiveProblem();
symbClassProblem.Name = "Mammography Classification Problem";
symbClassProblem.Description = "Mammography dataset imported from the UCI machine learning repository (http://archive.ics.uci.edu/ml/datasets/Mammographic+Mass)";
UCIInstanceProvider provider = new UCIInstanceProvider();
var instance = provider.GetDataDescriptors().Where(x => x.Name.Equals("Mammography, M. Elter, 2007")).Single();
var mammoData = (ClassificationProblemData)provider.LoadData(instance);
mammoData.TargetVariableParameter.Value = mammoData.TargetVariableParameter.ValidValues
.First(v => v.Value == "Severity");
mammoData.InputVariables.SetItemCheckedState(
mammoData.InputVariables.Single(x => x.Value == "BI-RADS"), false);
mammoData.InputVariables.SetItemCheckedState(
mammoData.InputVariables.Single(x => x.Value == "Age"), true);
mammoData.InputVariables.SetItemCheckedState(
mammoData.InputVariables.Single(x => x.Value == "Shape"), true);
mammoData.InputVariables.SetItemCheckedState(
mammoData.InputVariables.Single(x => x.Value == "Margin"), true);
mammoData.InputVariables.SetItemCheckedState(
mammoData.InputVariables.Single(x => x.Value == "Density"), true);
mammoData.InputVariables.SetItemCheckedState(
mammoData.InputVariables.Single(x => x.Value == "Severity"), false);
mammoData.TrainingPartition.Start = 0;
mammoData.TrainingPartition.End = 800;
mammoData.TestPartition.Start = 800;
mammoData.TestPartition.End = 961;
mammoData.Name = "Data imported from mammographic_masses.csv";
mammoData.Description = "Original dataset: http://archive.ics.uci.edu/ml/datasets/Mammographic+Mass, missing values have been replaced with median values.";
symbClassProblem.ProblemData = mammoData;
// configure grammar
var grammar = new TypeCoherentExpressionGrammar();
grammar.ConfigureAsDefaultClassificationGrammar();
grammar.Symbols.OfType<VariableCondition>().Single().Enabled = false;
var varSymbol = grammar.Symbols.OfType<Variable>().Where(x => !(x is LaggedVariable)).Single();
varSymbol.WeightMu = 1.0;
varSymbol.WeightSigma = 1.0;
varSymbol.WeightManipulatorMu = 0.0;
varSymbol.WeightManipulatorSigma = 0.05;
varSymbol.MultiplicativeWeightManipulatorSigma = 0.03;
var constSymbol = grammar.Symbols.OfType<Constant>().Single();
constSymbol.MaxValue = 20;
constSymbol.MinValue = -20;
constSymbol.ManipulatorMu = 0.0;
constSymbol.ManipulatorSigma = 1;
constSymbol.MultiplicativeManipulatorSigma = 0.03;
symbClassProblem.SymbolicExpressionTreeGrammar = grammar;
// configure remaining problem parameters
symbClassProblem.BestKnownQuality.Value = 0.0;
symbClassProblem.FitnessCalculationPartition.Start = 0;
symbClassProblem.FitnessCalculationPartition.End = 400;
symbClassProblem.ValidationPartition.Start = 400;
symbClassProblem.ValidationPartition.End = 800;
symbClassProblem.RelativeNumberOfEvaluatedSamples.Value = 1;
symbClassProblem.MaximumSymbolicExpressionTreeLength.Value = 100;
symbClassProblem.MaximumSymbolicExpressionTreeDepth.Value = 10;
symbClassProblem.MaximumFunctionDefinitions.Value = 0;
symbClassProblem.MaximumFunctionArguments.Value = 0;
symbClassProblem.EvaluatorParameter.Value = new SymbolicClassificationSingleObjectiveMeanSquaredErrorEvaluator();
#endregion
#region Algorithm Configuration
ga.Problem = symbClassProblem;
ga.Name = "Genetic Programming - Symbolic Classification";
ga.Description = "A standard genetic programming algorithm to solve a classification problem (Mammographic+Mass dataset)";
SamplesUtils.ConfigureGeneticAlgorithmParameters<TournamentSelector, SubtreeCrossover, MultiSymbolicExpressionTreeManipulator>(
ga, 1000, 1, 100, 0.15, 5
);
var mutator = (MultiSymbolicExpressionTreeManipulator)ga.Mutator;
mutator.Operators.OfType<FullTreeShaker>().Single().ShakingFactor = 0.1;
mutator.Operators.OfType<OnePointShaker>().Single().ShakingFactor = 1.0;
ga.Analyzer.Operators.SetItemCheckedState(
ga.Analyzer.Operators
.OfType<SymbolicClassificationSingleObjectiveOverfittingAnalyzer>()
.Single(), false);
ga.Analyzer.Operators.SetItemCheckedState(
ga.Analyzer.Operators
.OfType<SymbolicDataAnalysisAlleleFrequencyAnalyzer>()
.First(), false);
#endregion
return ga;
}
protected TypeCoherentExpressionGrammar(TypeCoherentExpressionGrammar original, Cloner cloner) : base(original, cloner) { }
protected TypeCoherentExpressionGrammar(TypeCoherentExpressionGrammar original, Cloner cloner) : base(original, cloner)
{
}
private GeneticAlgorithm CreateGpSymbolicRegressionSample() {
GeneticAlgorithm ga = new GeneticAlgorithm();
#region Problem Configuration
SymbolicRegressionSingleObjectiveProblem symbRegProblem = new SymbolicRegressionSingleObjectiveProblem();
symbRegProblem.Name = "Tower Symbolic Regression Problem";
symbRegProblem.Description = "Tower Dataset (downloaded from: http://www.symbolicregression.com/?q=towerProblem)";
RegressionRealWorldInstanceProvider provider = new RegressionRealWorldInstanceProvider();
var instance = provider.GetDataDescriptors().Where(x => x.Name.Equals("Tower")).Single();
var towerProblemData = (RegressionProblemData)provider.LoadData(instance);
towerProblemData.TargetVariableParameter.Value = towerProblemData.TargetVariableParameter.ValidValues
.First(v => v.Value == "towerResponse");
towerProblemData.InputVariables.SetItemCheckedState(
towerProblemData.InputVariables.Single(x => x.Value == "x1"), true);
towerProblemData.InputVariables.SetItemCheckedState(
towerProblemData.InputVariables.Single(x => x.Value == "x7"), false);
towerProblemData.InputVariables.SetItemCheckedState(
towerProblemData.InputVariables.Single(x => x.Value == "x11"), false);
towerProblemData.InputVariables.SetItemCheckedState(
towerProblemData.InputVariables.Single(x => x.Value == "x16"), false);
towerProblemData.InputVariables.SetItemCheckedState(
towerProblemData.InputVariables.Single(x => x.Value == "x21"), false);
towerProblemData.InputVariables.SetItemCheckedState(
towerProblemData.InputVariables.Single(x => x.Value == "x25"), false);
towerProblemData.InputVariables.SetItemCheckedState(
towerProblemData.InputVariables.Single(x => x.Value == "towerResponse"), false);
towerProblemData.TrainingPartition.Start = 0;
towerProblemData.TrainingPartition.End = 3136;
towerProblemData.TestPartition.Start = 3136;
towerProblemData.TestPartition.End = 4999;
towerProblemData.Name = "Data imported from towerData.txt";
towerProblemData.Description = "Chemical concentration at top of distillation tower, dataset downloaded from: http://vanillamodeling.com/realproblems.html, best R² achieved with nu-SVR = 0.97";
symbRegProblem.ProblemData = towerProblemData;
// configure grammar
var grammar = new TypeCoherentExpressionGrammar();
grammar.ConfigureAsDefaultRegressionGrammar();
grammar.Symbols.OfType<VariableCondition>().Single().InitialFrequency = 0.0;
var varSymbol = grammar.Symbols.OfType<Variable>().Where(x => !(x is LaggedVariable)).Single();
varSymbol.WeightMu = 1.0;
varSymbol.WeightSigma = 1.0;
varSymbol.WeightManipulatorMu = 0.0;
varSymbol.WeightManipulatorSigma = 0.05;
varSymbol.MultiplicativeWeightManipulatorSigma = 0.03;
var constSymbol = grammar.Symbols.OfType<Constant>().Single();
constSymbol.MaxValue = 20;
constSymbol.MinValue = -20;
constSymbol.ManipulatorMu = 0.0;
constSymbol.ManipulatorSigma = 1;
constSymbol.MultiplicativeManipulatorSigma = 0.03;
symbRegProblem.SymbolicExpressionTreeGrammar = grammar;
// configure remaining problem parameters
symbRegProblem.BestKnownQuality.Value = 0.97;
symbRegProblem.FitnessCalculationPartition.Start = 0;
symbRegProblem.FitnessCalculationPartition.End = 2300;
symbRegProblem.ValidationPartition.Start = 2300;
symbRegProblem.ValidationPartition.End = 3136;
symbRegProblem.RelativeNumberOfEvaluatedSamples.Value = 1;
symbRegProblem.MaximumSymbolicExpressionTreeLength.Value = 150;
symbRegProblem.MaximumSymbolicExpressionTreeDepth.Value = 12;
symbRegProblem.MaximumFunctionDefinitions.Value = 0;
symbRegProblem.MaximumFunctionArguments.Value = 0;
symbRegProblem.EvaluatorParameter.Value = new SymbolicRegressionSingleObjectivePearsonRSquaredEvaluator();
#endregion
#region Algorithm Configuration
ga.Problem = symbRegProblem;
ga.Name = "Genetic Programming - Symbolic Regression";
ga.Description = "A standard genetic programming algorithm to solve a symbolic regression problem (tower dataset)";
SamplesUtils.ConfigureGeneticAlgorithmParameters<TournamentSelector, SubtreeCrossover, MultiSymbolicExpressionTreeManipulator>(
ga, 1000, 1, 50, 0.15, 5);
var mutator = (MultiSymbolicExpressionTreeManipulator)ga.Mutator;
mutator.Operators.OfType<FullTreeShaker>().Single().ShakingFactor = 0.1;
mutator.Operators.OfType<OnePointShaker>().Single().ShakingFactor = 1.0;
ga.Analyzer.Operators.SetItemCheckedState(
ga.Analyzer.Operators
.OfType<SymbolicRegressionSingleObjectiveOverfittingAnalyzer>()
.Single(), false);
ga.Analyzer.Operators.SetItemCheckedState(
ga.Analyzer.Operators
.OfType<SymbolicDataAnalysisAlleleFrequencyAnalyzer>()
.First(), false);
#endregion
return ga;
}