private static ISymbolicRegressionSolution CreateSymbolicSolution(List <IRegressionModel> models, double nu, IRegressionProblemData problemData)
{
var symbModels = models.OfType <ISymbolicRegressionModel>();
var lowerLimit = symbModels.Min(m => m.LowerEstimationLimit);
var upperLimit = symbModels.Max(m => m.UpperEstimationLimit);
var interpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
var progRootNode = new ProgramRootSymbol().CreateTreeNode();
var startNode = new StartSymbol().CreateTreeNode();
var addNode = new Addition().CreateTreeNode();
var mulNode = new Multiplication().CreateTreeNode();
var scaleNode = (ConstantTreeNode) new Constant().CreateTreeNode(); // all models are scaled using the same nu
scaleNode.Value = nu;
foreach (var m in symbModels)
{
var relevantPart = m.SymbolicExpressionTree.Root.GetSubtree(0).GetSubtree(0); // skip root and start
addNode.AddSubtree((ISymbolicExpressionTreeNode)relevantPart.Clone());
}
mulNode.AddSubtree(addNode);
mulNode.AddSubtree(scaleNode);
startNode.AddSubtree(mulNode);
progRootNode.AddSubtree(startNode);
var t = new SymbolicExpressionTree(progRootNode);
var combinedModel = new SymbolicRegressionModel(problemData.TargetVariable, t, interpreter, lowerLimit, upperLimit);
var sol = new SymbolicRegressionSolution(combinedModel, problemData);
return(sol);
}
protected override void UpdateModel(ISymbolicExpressionTree tree)
{
var model = new SymbolicRegressionModel(Content.ProblemData.TargetVariable, tree, Content.Model.Interpreter, Content.Model.LowerEstimationLimit, Content.Model.UpperEstimationLimit);
model.Scale(Content.ProblemData);
Content.Model = model;
}
private IContent CreateModel(int idx)
{
idx -= 1;
var rfModel = Content.Model as RandomForestModel;
if (rfModel == null)
{
return(null);
}
var regProblemData = Content.ProblemData as IRegressionProblemData;
var classProblemData = Content.ProblemData as IClassificationProblemData;
if (idx < 0 || idx >= rfModel.NumberOfTrees)
{
return(null);
}
if (regProblemData != null)
{
var syModel = new SymbolicRegressionModel(regProblemData.TargetVariable, rfModel.ExtractTree(idx),
new SymbolicDataAnalysisExpressionTreeLinearInterpreter());
return(syModel.CreateRegressionSolution(regProblemData));
}
else if (classProblemData != null)
{
var syModel = new SymbolicDiscriminantFunctionClassificationModel(classProblemData.TargetVariable, rfModel.ExtractTree(idx),
new SymbolicDataAnalysisExpressionTreeLinearInterpreter(), new NormalDistributionCutPointsThresholdCalculator());
syModel.RecalculateModelParameters(classProblemData, classProblemData.TrainingIndices);
return(syModel.CreateClassificationSolution(classProblemData));
}
else
{
throw new InvalidProgramException();
}
}
public static ISymbolicRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData, string modelStructure, int maxIterations)
{
var parser = new InfixExpressionParser();
var tree = parser.Parse(modelStructure);
var simplifier = new SymbolicDataAnalysisExpressionTreeSimplifier();
if (!SymbolicRegressionConstantOptimizationEvaluator.CanOptimizeConstants(tree))
{
throw new ArgumentException("The optimizer does not support the specified model structure.");
}
var interpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, tree, problemData, problemData.TrainingIndices,
applyLinearScaling: false, maxIterations: maxIterations,
updateVariableWeights: false, updateConstantsInTree: true);
var scaledModel = new SymbolicRegressionModel(problemData.TargetVariable, tree, (ISymbolicDataAnalysisExpressionTreeInterpreter)interpreter.Clone());
scaledModel.Scale(problemData);
SymbolicRegressionSolution solution = new SymbolicRegressionSolution(scaledModel, (IRegressionProblemData)problemData.Clone());
solution.Model.Name = "Regression Model";
solution.Name = "Regression Solution";
return(solution);
}
public void CustomModelVariableImpactTest()
{
IRegressionProblemData problemData = CreateDefaultProblem();
ISymbolicExpressionTree tree = CreateCustomExpressionTree();
IRegressionModel model = new SymbolicRegressionModel(problemData.TargetVariable, tree, new SymbolicDataAnalysisExpressionTreeInterpreter());
IRegressionSolution solution = new RegressionSolution(model, (IRegressionProblemData)problemData.Clone());
Dictionary <string, double> expectedImpacts = GetExpectedValuesForCustomProblem();
CheckDefaultAsserts(solution, expectedImpacts);
}
protected override void Run(CancellationToken cancellationToken)
{
var interpreter = new SymbolicDataAnalysisExpressionTreeInterpreter();
var modelsWithComplexity = new List <(ISymbolicRegressionModel, int)>();
Stopwatch stopwatch = Stopwatch.StartNew();
RunFFX(out var basisFunctions, out var lambda, out var coeff, out var trainNMSE, out var testNMSE, out var intercept, out var elnetData);
stopwatch.Stop();
var runtime = stopwatch.ElapsedTicks / 10000000.0; // in seconds
if (Verbose)
{
var errorTable = NMSEGraph(coeff, lambda, trainNMSE, testNMSE);
Results.Add(new Result(errorTable.Name, errorTable.Description, errorTable));
var coeffTable = CoefficientGraph(coeff, lambda, elnetData.AllowedInputVariables, elnetData.Dataset);
Results.Add(new Result(coeffTable.Name, coeffTable.Description, coeffTable));
}
int complexity(double[] modelCoeffs) => modelCoeffs.Count(val => val != 0) + 1;
for (int row = 0; row < coeff.GetUpperBound(0); row++)
{
var coeffs = Utils.GetRow(coeff, row);
var numBasisFuncs = complexity(coeffs);
ISymbolicExpressionTree tree = Tree(basisFunctions, coeffs, intercept[row]);
ISymbolicRegressionModel model = new SymbolicRegressionModel(elnetData.TargetVariable, tree, interpreter);
modelsWithComplexity.Add((model, numBasisFuncs));
}
// calculate the pareto front
var paretoFront = Utils.NondominatedFilter(modelsWithComplexity.ToArray(), coeff, testNMSE, complexity);
modelsWithComplexity = modelsWithComplexity.Distinct(new Utils.SymbolicRegressionModelSameComplexity()).ToList();
var results = new ItemCollection <IResult>();
int modelIdx = 1;
foreach (var model in modelsWithComplexity)
{
results.Add(new Result("Model " + (modelIdx < 10 ? "0" + modelIdx : modelIdx.ToString()), model.Item1));
//results.Add(new ItemCollection<IResult>(3){
//new Result("Model Complexity", new IntValue(model.Item2)),
//new Result("Model Accuracy", new RegressionSolution(model.Item1, Problem.ProblemData))
//});
modelIdx++;
}
Results.Add(new Result("Pareto Front", "The Pareto Front of the Models. ", results));
if (FilePath != "")
{
SaveInFile(paretoFront, runtime, FilePath, ",");
}
}
private IContent CreateModel(int idx)
{
idx -= 1;
var rfModel = Content;
var rfClassModel = rfModel as IClassificationModel; // rfModel is always a IRegressionModel and a IClassificationModel
var targetVariable = rfClassModel.TargetVariable;
if (rfModel == null)
{
return(null);
}
if (idx < 0 || idx >= rfModel.NumberOfTrees)
{
return(null);
}
var syModel = new SymbolicRegressionModel(targetVariable, rfModel.ExtractTree(idx),
new SymbolicDataAnalysisExpressionTreeLinearInterpreter());
return(syModel);
}
protected override void UpdateModel(ISymbolicExpressionTree tree) {
var model = new SymbolicRegressionModel(Content.ProblemData.TargetVariable, tree, Content.Model.Interpreter, Content.Model.LowerEstimationLimit, Content.Model.UpperEstimationLimit);
model.Scale(Content.ProblemData);
Content.Model = model;
}
/// <summary>
/// Fits a model to the data by optimizing the numeric constants.
/// Model is specified as infix expression containing variable names and numbers.
/// The starting point for the numeric constants is initialized randomly if a random number generator is specified (~N(0,1)). Otherwise the user specified constants are
/// used as a starting point.
/// </summary>-
/// <param name="problemData">Training and test data</param>
/// <param name="modelStructure">The function as infix expression</param>
/// <param name="maxIterations">Number of constant optimization iterations (using Levenberg-Marquardt algorithm)</param>
/// <param name="random">Optional random number generator for random initialization of numeric constants.</param>
/// <returns></returns>
public static ISymbolicRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData, string modelStructure, int maxIterations, bool applyLinearScaling, IRandom rand = null)
{
var parser = new InfixExpressionParser();
var tree = parser.Parse(modelStructure);
// parser handles double and string variables equally by creating a VariableTreeNode
// post-process to replace VariableTreeNodes by FactorVariableTreeNodes for all string variables
var factorSymbol = new FactorVariable();
factorSymbol.VariableNames =
problemData.AllowedInputVariables.Where(name => problemData.Dataset.VariableHasType <string>(name));
factorSymbol.AllVariableNames = factorSymbol.VariableNames;
factorSymbol.VariableValues =
factorSymbol.VariableNames.Select(name =>
new KeyValuePair <string, Dictionary <string, int> >(name,
problemData.Dataset.GetReadOnlyStringValues(name).Distinct()
.Select((n, i) => Tuple.Create(n, i))
.ToDictionary(tup => tup.Item1, tup => tup.Item2)));
foreach (var parent in tree.IterateNodesPrefix().ToArray())
{
for (int i = 0; i < parent.SubtreeCount; i++)
{
var varChild = parent.GetSubtree(i) as VariableTreeNode;
var factorVarChild = parent.GetSubtree(i) as FactorVariableTreeNode;
if (varChild != null && factorSymbol.VariableNames.Contains(varChild.VariableName))
{
parent.RemoveSubtree(i);
var factorTreeNode = (FactorVariableTreeNode)factorSymbol.CreateTreeNode();
factorTreeNode.VariableName = varChild.VariableName;
factorTreeNode.Weights =
factorTreeNode.Symbol.GetVariableValues(factorTreeNode.VariableName).Select(_ => 1.0).ToArray();
// weight = 1.0 for each value
parent.InsertSubtree(i, factorTreeNode);
}
else if (factorVarChild != null && factorSymbol.VariableNames.Contains(factorVarChild.VariableName))
{
if (factorSymbol.GetVariableValues(factorVarChild.VariableName).Count() != factorVarChild.Weights.Length)
{
throw new ArgumentException(
string.Format("Factor variable {0} needs exactly {1} weights",
factorVarChild.VariableName,
factorSymbol.GetVariableValues(factorVarChild.VariableName).Count()));
}
parent.RemoveSubtree(i);
var factorTreeNode = (FactorVariableTreeNode)factorSymbol.CreateTreeNode();
factorTreeNode.VariableName = factorVarChild.VariableName;
factorTreeNode.Weights = factorVarChild.Weights;
parent.InsertSubtree(i, factorTreeNode);
}
}
}
if (!SymbolicRegressionConstantOptimizationEvaluator.CanOptimizeConstants(tree))
{
throw new ArgumentException("The optimizer does not support the specified model structure.");
}
// initialize constants randomly
if (rand != null)
{
foreach (var node in tree.IterateNodesPrefix().OfType <ConstantTreeNode>())
{
double f = Math.Exp(NormalDistributedRandom.NextDouble(rand, 0, 1));
double s = rand.NextDouble() < 0.5 ? -1 : 1;
node.Value = s * node.Value * f;
}
}
var interpreter = new SymbolicDataAnalysisExpressionTreeLinearInterpreter();
SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(interpreter, tree, problemData, problemData.TrainingIndices,
applyLinearScaling: applyLinearScaling, maxIterations: maxIterations,
updateVariableWeights: false, updateConstantsInTree: true);
var model = new SymbolicRegressionModel(problemData.TargetVariable, tree, (ISymbolicDataAnalysisExpressionTreeInterpreter)interpreter.Clone());
if (applyLinearScaling)
{
model.Scale(problemData);
}
SymbolicRegressionSolution solution = new SymbolicRegressionSolution(model, (IRegressionProblemData)problemData.Clone());
solution.Model.Name = "Regression Model";
solution.Name = "Regression Solution";
return(solution);
}