Free cookie consent management tool by TermsFeed Policy Generator

Changeset 16507


Ignore:
Timestamp:
01/06/19 18:03:15 (6 years ago)
Author:
mkommend
Message:

#2974: First stable version of new CoOp.

Location:
branches/2974_Constants_Optimization
Files:
1 added
7 edited
1 copied
1 moved

Legend:

Unmodified
Added
Removed

  • branches/2974_Constants_Optimization/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/ConstantsOptimization/AutoDiffConverter.cs

    r16501 r16507  
    2727using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
    2828
    29 namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
    30   public class TreeToAutoDiffTermConverter {
    31     public delegate double ParametricFunction(double[] vars, double[] @params);
    32 
    33     public delegate Tuple<double[], double> ParametricFunctionGradient(double[] vars, double[] @params);
    34 
    35     #region helper class
    36     public class DataForVariable {
    37       public readonly string variableName;
    38       public readonly string variableValue; // for factor vars
    39       public readonly int lag;
    40 
    41       public DataForVariable(string varName, string varValue, int lag) {
    42         this.variableName = varName;
    43         this.variableValue = varValue;
    44         this.lag = lag;
    45       }
    46 
    47       public override bool Equals(object obj) {
    48         var other = obj as DataForVariable;
    49         if (other == null) return false;
    50         return other.variableName.Equals(this.variableName) &&
    51                other.variableValue.Equals(this.variableValue) &&
    52                other.lag == this.lag;
    53       }
    54 
    55       public override int GetHashCode() {
    56         return variableName.GetHashCode() ^ variableValue.GetHashCode() ^ lag;
    57       }
    58     }
    59     #endregion
    60 
    61     #region derivations of functions
    62     // create function factory for arctangent
    63     private static readonly Func<Term, UnaryFunc> arctan = UnaryFunc.Factory(
    64       eval: Math.Atan,
    65       diff: x => 1 / (1 + x * x));
    66 
    67     private static readonly Func<Term, UnaryFunc> sin = UnaryFunc.Factory(
    68       eval: Math.Sin,
    69       diff: Math.Cos);
    70 
    71     private static readonly Func<Term, UnaryFunc> cos = UnaryFunc.Factory(
    72       eval: Math.Cos,
    73       diff: x => -Math.Sin(x));
    74 
    75     private static readonly Func<Term, UnaryFunc> tan = UnaryFunc.Factory(
    76       eval: Math.Tan,
    77       diff: x => 1 + Math.Tan(x) * Math.Tan(x));
    78 
    79     private static readonly Func<Term, UnaryFunc> erf = UnaryFunc.Factory(
    80       eval: alglib.errorfunction,
    81       diff: x => 2.0 * Math.Exp(-(x * x)) / Math.Sqrt(Math.PI));
    82 
    83     private static readonly Func<Term, UnaryFunc> norm = UnaryFunc.Factory(
    84       eval: alglib.normaldistribution,
    85       diff: x => -(Math.Exp(-(x * x)) * Math.Sqrt(Math.Exp(x * x)) * x) / Math.Sqrt(2 * Math.PI));
    86 
    87     private static readonly Func<Term, UnaryFunc> abs = UnaryFunc.Factory(
    88       eval: Math.Abs,
    89       diff: x => Math.Sign(x)
    90       );
    91 
    92     #endregion
    93 
    94     public static bool TryConvertToAutoDiff(ISymbolicExpressionTree tree, bool makeVariableWeightsVariable, bool addLinearScalingTerms,
    95       out List<DataForVariable> parameters, out double[] initialConstants,
    96       out ParametricFunction func,
    97       out ParametricFunctionGradient func_grad) {
    98 
     29namespace HeuristicLab.Problems.DataAnalysis.Symbolic.ConstantsOptimization{
     30  public class AutoDiffConverter {
     31
     32    /// <summary>
     33    /// Converts a symbolic expression tree into a parametetric AutoDiff term.
     34    /// </summary>
     35    /// <param name="tree">The tree the should be converted.</param>
     36    /// <param name="addLinearScalingTerms">A flag that determines whether linear scaling terms should be added to the parametric term.</param>
     37    /// <param name="numericNodes">The nodes that contain numeric coefficents that should be added as variables in the term.</param>
     38    /// <param name="variableData">The variable information that is used to create parameters in the term.</param>
     39    /// <param name="autoDiffTerm">The resulting parametric AutoDiff term.</param>
     40    /// <returns>A flag to see if the conversion has succeeded.</returns>
     41    public static bool TryConvertToAutoDiff(ISymbolicExpressionTree tree, bool addLinearScalingTerms,
     42      IEnumerable<ISymbolicExpressionTreeNode> numericNodes, IEnumerable<VariableData> variableData,
     43      out IParametricCompiledTerm autoDiffTerm) {
    9944      // use a transformator object which holds the state (variable list, parameter list, ...) for recursive transformation of the tree
    100       var transformator = new TreeToAutoDiffTermConverter(makeVariableWeightsVariable);
     45      var transformator = new AutoDiffConverter(numericNodes, variableData);
    10146      AutoDiff.Term term;
    102       try {
    103         term = transformator.ConvertToAutoDiff(tree.Root.GetSubtree(0));
    104 
    105         if (addLinearScalingTerms) {
    106           // scaling variables α, β are given at the beginning of the parameter vector
    107           var alpha = new AutoDiff.Variable();
    108           var beta = new AutoDiff.Variable();
    109           transformator.variables.Insert(0, alpha);
    110           transformator.variables.Insert(0, beta);
    111 
    112           term = term * alpha + beta;
    113         }
    114 
    115         var parameterEntries = transformator.parameters.ToArray(); // guarantee same order for keys and values
    116         var compiledTerm = term.Compile(transformator.variables.ToArray(),
    117           parameterEntries.Select(kvp => kvp.Value).ToArray());
    118 
    119         parameters = new List<DataForVariable>(parameterEntries.Select(kvp => kvp.Key));
    120         initialConstants = transformator.initialConstants.ToArray();
    121         func = (vars, @params) => compiledTerm.Evaluate(vars, @params);
    122         func_grad = (vars, @params) => compiledTerm.Differentiate(vars, @params);
    123         return true;
    124       } catch (ConversionException) {
    125         parameters = null;
    126         initialConstants = null;
    127         func = null;
    128         func_grad = null;
    129       }
    130       return false;
    131     }
    132 
    133     public static bool TryConvertToAutoDiff(ISymbolicExpressionTree tree, bool addLinearScalingTerms, IEnumerable<DataForVariable> variables,
    134       out IParametricCompiledTerm autoDiffTerm, out double[] initialConstants) {
    135       // use a transformator object which holds the state (variable list, parameter list, ...) for recursive transformation of the tree
    136       //TODO change ctor
    137       var transformator = new TreeToAutoDiffTermConverter(true);
    138       var parameters = new AutoDiff.Variable[variables.Count()];
    139 
    140       int i = 0;
    141       foreach(var variable in variables) {
    142         var autoDiffVar = new AutoDiff.Variable();
    143         transformator.parameters.Add(variable, autoDiffVar);
    144         parameters[i] = autoDiffVar;
    145         i++;
    146       }
    147 
    148       AutoDiff.Term term;
     47
    14948      try {
    15049        term = transformator.ConvertToAutoDiff(tree.Root.GetSubtree(0));
    … …  
    15857          transformator.variables.Add(alpha);
    15958          transformator.variables.Add(beta);
    160 
    161           transformator.initialConstants.Add(1.0);
    162           transformator.initialConstants.Add(0.0);
    163         }
    164 
    165         var compiledTerm = term.Compile(transformator.variables.ToArray(), parameters);
     59        }
     60        var compiledTerm = term.Compile(transformator.variables.ToArray(), transformator.parameters.Values.ToArray());
    16661        autoDiffTerm = compiledTerm;
    167         initialConstants = transformator.initialConstants.ToArray();
    168 
    16962        return true;
    17063      } catch (ConversionException) {
    17164        autoDiffTerm = null;
    172         initialConstants = null;
    17365      }
    17466      return false;
    … …  
    17668
    17769    // state for recursive transformation of trees
    178     private readonly List<double> initialConstants;
    179     private readonly Dictionary<DataForVariable, AutoDiff.Variable> parameters;
     70    private readonly HashSet<ISymbolicExpressionTreeNode> nodesForOptimization;
     71    private readonly Dictionary<VariableData, AutoDiff.Variable> parameters;
    18072    private readonly List<AutoDiff.Variable> variables;
    181     private readonly bool makeVariableWeightsVariable;
    182 
    183     private TreeToAutoDiffTermConverter(bool makeVariableWeightsVariable) {
    184       this.makeVariableWeightsVariable = makeVariableWeightsVariable;
    185       this.initialConstants = new List<double>();
    186       this.parameters = new Dictionary<DataForVariable, AutoDiff.Variable>();
     73
     74    private AutoDiffConverter(IEnumerable<ISymbolicExpressionTreeNode> nodesForOptimization, IEnumerable<VariableData> variableData) {
     75      this.nodesForOptimization = new HashSet<ISymbolicExpressionTreeNode>(nodesForOptimization);
     76      this.parameters = variableData.ToDictionary(k => k, v => new AutoDiff.Variable());
    18777      this.variables = new List<AutoDiff.Variable>();
    18878    }
    … …  
    19080    private AutoDiff.Term ConvertToAutoDiff(ISymbolicExpressionTreeNode node) {
    19181      if (node.Symbol is Constant) {
    192         initialConstants.Add(((ConstantTreeNode)node).Value);
    193         var var = new AutoDiff.Variable();
    194         variables.Add(var);
    195         return var;
     82        var constantNode = node as ConstantTreeNode;
     83        var value = constantNode.Value;
     84        if (nodesForOptimization.Contains(node)) {
     85          AutoDiff.Variable var = new AutoDiff.Variable();
     86          variables.Add(var);
     87          return var;
     88        } else {
     89          return value;
     90        }
    19691      }
    19792      if (node.Symbol is Variable || node.Symbol is BinaryFactorVariable) {
    … …  
    20095        // factor variable values are only 0 or 1 and set in x accordingly
    20196        var varValue = factorVarNode != null ? factorVarNode.VariableValue : string.Empty;
    202         var par = FindOrCreateParameter(parameters, varNode.VariableName, varValue);
    203 
    204         if (makeVariableWeightsVariable) {
    205           initialConstants.Add(varNode.Weight);
    206           var w = new AutoDiff.Variable();
    207           variables.Add(w);
    208           return AutoDiff.TermBuilder.Product(w, par);
     97        var data = new VariableData(varNode.VariableName, varValue, 0);
     98        var par = parameters[data];
     99        var value = varNode.Weight;
     100
     101        if (nodesForOptimization.Contains(node)) {
     102          AutoDiff.Variable var = new AutoDiff.Variable();
     103          variables.Add(var);
     104          return AutoDiff.TermBuilder.Product(var, par);
    209105        } else {
    210           return varNode.Weight * par;
     106          return AutoDiff.TermBuilder.Product(value, par);
    211107        }
    212108      }
    … …  
    215111        var products = new List<Term>();
    216112        foreach (var variableValue in factorVarNode.Symbol.GetVariableValues(factorVarNode.VariableName)) {
    217           var par = FindOrCreateParameter(parameters, factorVarNode.VariableName, variableValue);
    218 
    219           initialConstants.Add(factorVarNode.GetValue(variableValue));
    220           var wVar = new AutoDiff.Variable();
    221           variables.Add(wVar);
    222 
    223           products.Add(AutoDiff.TermBuilder.Product(wVar, par));
     113          var data = new VariableData(factorVarNode.VariableName, variableValue, 0);
     114          var par = parameters[data];
     115          var value = factorVarNode.GetValue(variableValue);
     116
     117          if (nodesForOptimization.Contains(node)) {
     118            var wVar = new AutoDiff.Variable();
     119            variables.Add(wVar);
     120
     121            products.Add(AutoDiff.TermBuilder.Product(wVar, par));
     122          } else {
     123            products.Add(AutoDiff.TermBuilder.Product(value, par));
     124          }
    224125        }
    225126        return AutoDiff.TermBuilder.Sum(products);
    … …  
    227128      if (node.Symbol is LaggedVariable) {
    228129        var varNode = node as LaggedVariableTreeNode;
    229         var par = FindOrCreateParameter(parameters, varNode.VariableName, string.Empty, varNode.Lag);
    230 
    231         if (makeVariableWeightsVariable) {
    232           initialConstants.Add(varNode.Weight);
    233           var w = new AutoDiff.Variable();
    234           variables.Add(w);
    235           return AutoDiff.TermBuilder.Product(w, par);
     130        var data = new VariableData(varNode.VariableName, string.Empty, varNode.Lag);
     131        var par = parameters[data];
     132        var value = varNode.Weight;
     133
     134        if (nodesForOptimization.Contains(node)) {
     135          AutoDiff.Variable var = new AutoDiff.Variable();
     136          variables.Add(var);
     137          return AutoDiff.TermBuilder.Product(var, par);
    236138        } else {
    237           return varNode.Weight * par;
    238         }
     139          return AutoDiff.TermBuilder.Product(value, par);
     140        }
     141
    239142      }
    240143      if (node.Symbol is Addition) {
    … …  
    330233    }
    331234
    332 
    333     // for each factor variable value we need a parameter which represents a binary indicator for that variable & value combination
    334     // each binary indicator is only necessary once. So we only create a parameter if this combination is not yet available
    335     private static Term FindOrCreateParameter(Dictionary<DataForVariable, AutoDiff.Variable> parameters,
    336       string varName, string varValue = "", int lag = 0) {
    337       var data = new DataForVariable(varName, varValue, lag);
    338 
    339       AutoDiff.Variable par = null;
    340       if (!parameters.TryGetValue(data, out par)) {
    341         // not found -> create new parameter and entries in names and values lists
    342         par = new AutoDiff.Variable();
    343         parameters.Add(data, par);
    344       }
    345       return par;
    346     }
     235    #region derivations of functions
     236    // create function factory for arctangent
     237    private static readonly Func<Term, UnaryFunc> arctan = UnaryFunc.Factory(
     238      eval: Math.Atan,
     239      diff: x => 1 / (1 + x * x));
     240
     241    private static readonly Func<Term, UnaryFunc> sin = UnaryFunc.Factory(
     242      eval: Math.Sin,
     243      diff: Math.Cos);
     244
     245    private static readonly Func<Term, UnaryFunc> cos = UnaryFunc.Factory(
     246      eval: Math.Cos,
     247      diff: x => -Math.Sin(x));
     248
     249    private static readonly Func<Term, UnaryFunc> tan = UnaryFunc.Factory(
     250      eval: Math.Tan,
     251      diff: x => 1 + Math.Tan(x) * Math.Tan(x));
     252
     253    private static readonly Func<Term, UnaryFunc> erf = UnaryFunc.Factory(
     254      eval: alglib.errorfunction,
     255      diff: x => 2.0 * Math.Exp(-(x * x)) / Math.Sqrt(Math.PI));
     256
     257    private static readonly Func<Term, UnaryFunc> norm = UnaryFunc.Factory(
     258      eval: alglib.normaldistribution,
     259      diff: x => -(Math.Exp(-(x * x)) * Math.Sqrt(Math.Exp(x * x)) * x) / Math.Sqrt(2 * Math.PI));
     260
     261    private static readonly Func<Term, UnaryFunc> abs = UnaryFunc.Factory(
     262      eval: Math.Abs,
     263      diff: x => Math.Sign(x)
     264      );
     265
     266    #endregion
     267
    347268
    348269    public static bool IsCompatible(ISymbolicExpressionTree tree) {
    … …  
    379300    [Serializable]
    380301    public class ConversionException : Exception {
    381 
    382       public ConversionException() {
    383       }
    384 
    385       public ConversionException(string message) : base(message) {
    386       }
    387 
    388       public ConversionException(string message, Exception inner) : base(message, inner) {
    389       }
    390 
     302      public ConversionException() { }
     303      public ConversionException(string message) : base(message) { }
     304      public ConversionException(string message, Exception inner) : base(message, inner) { }
    391305      protected ConversionException(
    392306        SerializationInfo info,

  • branches/2974_Constants_Optimization/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/ConstantsOptimization/IConstantsOptimizer.cs

    r16500 r16507  
    2323using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
    2424
    25 namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
     25namespace HeuristicLab.Problems.DataAnalysis.Symbolic.ConstantsOptimization {
    2626  public interface IConstantsOptimizer {
    2727    bool ApplyLinearScaling { get; set; }

  • branches/2974_Constants_Optimization/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/ConstantsOptimization/LMConstantsOptimizer.cs

    r16500 r16507  
    2525using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
    2626
    27 namespace HeuristicLab.Problems.DataAnalysis.Symbolic {
     27namespace HeuristicLab.Problems.DataAnalysis.Symbolic.ConstantsOptimization {
    2828  public class LMConstantsOptimizer {
    29     private bool ApplyLinearScaling { get; set; }
    30     private int MaximumIterations { get; set; }
    3129
    32     public LMConstantsOptimizer() {
     30    private LMConstantsOptimizer() { }
    3331
     32    /// <summary>
     33    /// Method to determine whether the numeric constants of the tree can be optimized. This depends primarily on the symbols occuring in the tree.
     34    /// </summary>
     35    /// <param name="tree">The tree that should be analyzed</param>
     36    /// <returns>A flag indicating whether the numeric constants of the tree can be optimized</returns>
     37    public static bool CanOptimizeConstants(ISymbolicExpressionTree tree) {
     38      return AutoDiffConverter.IsCompatible(tree);
    3439    }
     40
    3541    /// <summary>
    36     ///
     42    /// Optimizes the numeric constants in a symbolic expression tree in place.
    3743    /// </summary>
    38     /// <param name="tree">The tree for which the constants are optimized.</param>
    39     /// <param name="nodes">The nodes which should be adapted. The nodes must be the same (reference) as the ones used in the tree and the double values specify the inita</param>
    40     /// <param name="x">The input data.</param>
    41     /// <param name="y">The targer date.</param>
    42     /// <param name="applyLinearScaling">Flag that determines whether linear scaling nodes should be added during the optimization.</param>
    43     /// <returns> Fit of the symbolic expression tree in terms of ... </returns>
    44     public static double OptimizeConstants(ISymbolicExpressionTree tree,IRegressionProblemData problemData, IEnumerable<int> rows, bool applyLinearScaling, int maxIterations = 10) {
    45       //if (tree == null) throw new ArgumentNullException("tree");
    46       //if (nodes == null) throw new ArgumentNullException("nodes");
    47       //if (initialConstants == null) throw new ArgumentNullException("intitialConstants");
    48       //if (problemData == null) throw new ArgumentNullException("problemData");
     44    /// <param name="tree">The tree for which the constants should be optimized</param>
     45    /// <param name="dataset">The dataset containing the data.</param>
     46    /// <param name="targetVariable">The target variable name.</param>
     47    /// <param name="rows">The rows for which the data should be extracted.</param>
     48    /// <param name="applyLinearScaling">A flag to determine whether linear scaling should be applied during the optimization</param>
     49    /// <param name="maxIterations">The maximum number of iterations of the Levenberg-Marquard algorithm.</param>
     50    /// <returns></returns>
     51    public static double OptimizeConstants(ISymbolicExpressionTree tree,
     52      IDataset dataset, string targetVariable, IEnumerable<int> rows,
     53      bool applyLinearScaling, int maxIterations = 10) {
     54      if (tree == null) throw new ArgumentNullException("tree");
     55      if (dataset == null) throw new ArgumentNullException("dataset");
     56      if (!dataset.ContainsVariable(targetVariable)) throw new ArgumentException("The dataset does not contain the provided target variable.");
    4957
    50       //if (!nodes.Any()) return 0;
    51 
    52       var ds = problemData.Dataset;
    53       var variables = ConstantsOptimization.Util.GenerateVariables(ds);
    54       var laggedVariables = ConstantsOptimization.Util.ExtractLaggedVariables(tree);
    55       var allVariables = variables.Union(laggedVariables);
    56 
    57       double[,] x = ConstantsOptimization.Util.ExtractData(ds, allVariables, rows);
    58       double[] y = ds.GetDoubleValues(problemData.TargetVariable, rows).ToArray();
    59 
    60       double[] initialConstants;
     58      var allVariables = Util.ExtractVariables(tree);
     59      var numericNodes = Util.ExtractNumericNodes(tree);
    6160
    6261      AutoDiff.IParametricCompiledTerm term;
    63       if (!TreeToAutoDiffTermConverter.TryConvertToAutoDiff(tree, applyLinearScaling, allVariables, out term, out initialConstants))
    64         throw new NotSupportedException("Could not optimize constants of symbolic expression tree due to not supported symbols used in the tree.");
     62      if (!AutoDiffConverter.TryConvertToAutoDiff(tree, applyLinearScaling, numericNodes, allVariables, out term))
     63        throw new NotSupportedException("Could not convert symbolic expression tree to an AutoDiff term due to not supported symbols used in the tree.");
    6564
    66       var constants = (double[])initialConstants.Clone();
     65      //Variables of the symbolic expression tree correspond to parameters in the term
     66      //Hence if no parameters are present no variables occur in the tree and the R² = 0
     67      if (term.Parameters.Count == 0) return 0.0;
     68
     69      var initialConstants = Util.ExtractConstants(numericNodes, applyLinearScaling);
     70      double[] constants;
     71      double[,] x = Util.ExtractData(dataset, allVariables, rows);
     72      double[] y = dataset.GetDoubleValues(targetVariable, rows).ToArray();
     73
     74      var result = OptimizeConstants(term, initialConstants, x, y, maxIterations, out constants);
     75      if (result != 0.0 && constants.Length != 0)
     76        Util.UpdateConstants(numericNodes, constants);
     77
     78      return result;
     79    }
     80
     81    /// <summary>
     82    /// Optimizes the numeric coefficents of an AutoDiff Term using the Levenberg-Marquard algorithm.
     83    /// </summary>
     84    /// <param name="term">The AutoDiff term for which the numeric coefficients should be optimized.</param>
     85    /// <param name="initialConstants">The starting values for the numeric coefficients.</param>
     86    /// <param name="x">The input data for the optimization.</param>
     87    /// <param name="y">The target values for the optimization.</param>
     88    /// <param name="maxIterations">The maximum number of iterations of the Levenberg-Marquard</param>
     89    /// <param name="constants">The opitmized constants.</param>
     90    /// <param name="LM_IterationCallback">An optional callback for detailed analysis that is called in each algorithm iteration.</param>
     91    /// <returns>The R² of the term evaluated on the input data x and the target data y using the optimized constants</returns>
     92    public static double OptimizeConstants(AutoDiff.IParametricCompiledTerm term, double[] initialConstants, double[,] x, double[] y,
     93      int maxIterations, out double[] constants, Action<double[], double, object> LM_IterationCallback = null) {
     94
     95      if (term.Parameters.Count == 0) {
     96        constants = new double[0];
     97        return 0.0;
     98      }
     99
     100      var optimizedConstants = (double[])initialConstants.Clone();
     101      int numberOfRows = x.GetLength(0);
     102      int numberOfColumns = x.GetLength(1);
     103      int numberOfConstants = optimizedConstants.Length;
     104
    67105      alglib.lsfitstate state;
    68106      alglib.lsfitreport rep;
     107      alglib.ndimensional_rep xrep = (p, f, obj) => LM_IterationCallback(p, f, obj);
    69108      int retVal;
    70109
    71       int numberOfRows = x.GetLength(0);
    72       int numberOfColumns = x.GetLength(1);
    73       int numberOfConstants = constants.Length;
    74 
    75110      try {
    76         alglib.lsfitcreatefg(x, y, constants, numberOfRows, numberOfColumns, numberOfConstants, cheapfg: false, state: out state);
     111        alglib.lsfitcreatefg(x, y, optimizedConstants, numberOfRows, numberOfColumns, numberOfConstants, cheapfg: false, state: out state);
    77112        alglib.lsfitsetcond(state, 0.0, 0.0, maxIterations);
    78         //alglib.lsfitsetgradientcheck(state, 0.001);
    79         alglib.lsfitfit(state, Evaluate, EvaluateGradient, null, term);
    80         alglib.lsfitresults(state, out retVal, out constants, out rep);
     113        alglib.lsfitsetxrep(state, LM_IterationCallback != null);
     114        alglib.lsfitfit(state, Evaluate, EvaluateGradient, xrep, term);
     115        alglib.lsfitresults(state, out retVal, out optimizedConstants, out rep);
    81116      } catch (ArithmeticException) {
     117        constants = new double[0];
    82118        return double.NaN;
    83119      } catch (alglib.alglibexception) {
     120        constants = new double[0];
    84121        return double.NaN;
    85122      }
    86123
    87       ConstantsOptimization.Util.UpdateConstants(tree, constants);
    88 
     124      constants = optimizedConstants;
    89125      return rep.r2;
    90126    }
     127
    91128
    92129    private static void Evaluate(double[] c, double[] x, ref double fx, object o) {
    … …  
    101138      Array.Copy(result.Item1, grad, grad.Length);
    102139    }
    103 
    104     public static bool CanOptimizeConstants(ISymbolicExpressionTree tree) {
    105       return TreeToAutoDiffTermConverter.IsCompatible(tree);
    106     }
    107140  }
    108141}

  • branches/2974_Constants_Optimization/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/ConstantsOptimization/Util.cs

    r16500 r16507  
    2525using HeuristicLab.Common;
    2626using HeuristicLab.Encodings.SymbolicExpressionTreeEncoding;
    27 using static HeuristicLab.Problems.DataAnalysis.Symbolic.TreeToAutoDiffTermConverter;
    2827
    2928namespace HeuristicLab.Problems.DataAnalysis.Symbolic.ConstantsOptimization {
    3029  public static class Util {
    31 
    32     public static double[,] ExtractData(IDataset dataset, IEnumerable<DataForVariable> variables, IEnumerable<int> rows) {
    33       var x = new double[rows.Count(), variables.Count()];
    34 
    35       int row = 0;
    36       foreach (var r in rows) {
    37         int col = 0;
    38         foreach (var variable in variables) {
    39           if (dataset.VariableHasType<double>(variable.variableName)) {
    40             x[row, col] = dataset.GetDoubleValue(variable.variableName, r + variable.lag);
    41           } else if (dataset.VariableHasType<string>(variable.variableName)) {
    42             x[row, col] = dataset.GetStringValue(variable.variableName, r) == variable.variableValue ? 1 : 0;
    43           } else throw new InvalidProgramException("found a variable of unknown type");
    44           col++;
    45         }
    46         row++;
    47       }
    48       return x;
    49     }
    50 
    51     public static List<DataForVariable> GenerateVariables(IDataset dataset) {
    52       var variables = new List<DataForVariable>();
    53       foreach (var doubleVariable in dataset.DoubleVariables) {
    54         var data = new DataForVariable(doubleVariable, string.Empty, 0);
    55         variables.Add(data);
    56       }
    57 
    58       foreach (var stringVariable in dataset.StringVariables) {
    59         foreach (var stringValue in dataset.GetStringValues(stringVariable).Distinct()) {
    60           var data = new DataForVariable(stringVariable, stringValue, 0);
     30    /// <summary>
     31    /// Extracts all variable information in a symbolic expression tree. The variable information is necessary to convert a tree in an AutoDiff term.
     32    /// </summary>
     33    /// <param name="tree">The tree referencing the variables.</param>
     34    /// <returns>The data for variables occuring in the tree.</returns>
     35    public static List<VariableData> ExtractVariables(ISymbolicExpressionTree tree) {
     36      if (tree == null) throw new ArgumentNullException("tree");
     37
     38      var variables = new HashSet<VariableData>();
     39      foreach (var node in tree.IterateNodesPrefix().OfType<IVariableTreeNode>()) {
     40        string variableName = node.VariableName;
     41        int lag = 0;
     42        var laggedNode = node as ILaggedTreeNode;
     43        if (laggedNode != null) lag = laggedNode.Lag;
     44
     45
     46        var factorNode = node as FactorVariableTreeNode;
     47        if (factorNode != null) {
     48          foreach (var factorValue in factorNode.Symbol.GetVariableValues(variableName)) {
     49            var data = new VariableData(variableName, factorValue, lag);
     50            variables.Add(data);
     51          }
     52        } else {
     53          var data = new VariableData(variableName, string.Empty, lag);
    6154          variables.Add(data);
    6255        }
    6356      }
    64       return variables;
    65     }
    66 
    67     public static List<DataForVariable> ExtractLaggedVariables(ISymbolicExpressionTree tree) {
    68       var variables = new HashSet<DataForVariable>();
    69       foreach (var laggedNode in tree.IterateNodesPrefix().OfType<ILaggedTreeNode>()) {
    70         var laggedVariableTreeNode = laggedNode as LaggedVariableTreeNode;
    71         if (laggedVariableTreeNode != null) {
    72           var data = new DataForVariable(laggedVariableTreeNode.VariableName, string.Empty, laggedVariableTreeNode.Lag);
    73           if (!variables.Contains(data)) variables.Add(data);
    74         }
    75       }
    7657      return variables.ToList();
    7758    }
    78 
    79     public static double[] ExtractConstants(ISymbolicExpressionTree tree) {
     59    /// <summary>
     60    /// Extract the necessary date for constants optimization with AutoDiff
     61    /// </summary>
     62    /// <param name="dataset">The dataset holding the data.</param>
     63    /// <param name="variables">The variables for which the data from the dataset should be extracted.</param>
     64    /// <param name="rows">The rows for which the data should be extracted.</param>
     65    /// <returns>A two-dimensiona double array containing the input data.</returns>
     66    public static double[,] ExtractData(IDataset dataset, IEnumerable<VariableData> variables, IEnumerable<int> rows) {
     67      if (dataset == null) throw new ArgumentNullException("dataset");
     68      if (variables == null) throw new ArgumentNullException("variables");
     69      if (rows == null) throw new ArgumentNullException("rows");
     70
     71      var x = new double[rows.Count(), variables.Count()];
     72
     73      int col = 0;
     74      foreach (var variable in variables) {
     75        if (dataset.VariableHasType<double>(variable.variableName)) {
     76          IEnumerable<double> values;
     77          if (variable.lag == 0)
     78            values = dataset.GetDoubleValues(variable.variableName, rows);
     79          else
     80            values = dataset.GetDoubleValues(variable.variableName, rows.Select(r => r + variable.lag));
     81
     82          int row = 0;
     83          foreach (var value in values) {
     84            x[row, col] = value;
     85            row++;
     86          }
     87        } else if (dataset.VariableHasType<string>(variable.variableName)) {
     88          var values = dataset.GetStringValues(variable.variableName, rows);
     89
     90          int row = 0;
     91          foreach (var value in values) {
     92            x[row, col] = value == variable.variableValue ? 1 : 0; ;
     93            row++;
     94          }
     95        } else throw new NotSupportedException("found a variable of unknown type");
     96        col++;
     97      }
     98
     99      return x;
     100    }
     101
     102    /// <summary>
     103    /// Extracts all numeric nodes from a symbolic expression tree that can be optimized by the constants optimization
     104    /// </summary>
     105    /// <param name="tree">The tree from which the numeric nodes should be extracted.</param>
     106    /// <returns>A list containing all nodes with numeric coefficients.</returns>
     107    public static List<ISymbolicExpressionTreeNode> ExtractNumericNodes(ISymbolicExpressionTree tree) {
     108      if (tree == null) throw new ArgumentNullException("tree");
     109
     110      var nodes = new List<ISymbolicExpressionTreeNode>();
     111      foreach (var node in tree.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>()) {
     112        ConstantTreeNode constantTreeNode = node as ConstantTreeNode;
     113        VariableTreeNodeBase variableTreeNodeBase = node as VariableTreeNodeBase;
     114        FactorVariableTreeNode factorVarTreeNode = node as FactorVariableTreeNode;
     115        if (constantTreeNode != null) nodes.Add(constantTreeNode);
     116        else if (variableTreeNodeBase != null) nodes.Add(variableTreeNodeBase);
     117        else if (factorVarTreeNode != null) nodes.Add(variableTreeNodeBase);
     118        else throw new NotSupportedException(string.Format("Terminal nodes of type {0} are not supported.", node.GetType().GetPrettyName()));
     119      }
     120      return nodes;
     121    }
     122
     123    /// <summary>
     124    /// Extracts all numeric constants from a symbolic expression tree.
     125    /// </summary>
     126    /// <param name="tree">The tree from which the numeric constants should be extracted.</param>
     127    /// <param name="addLinearScalingConstants">Flag to determine whether constants for linear scaling have to be added at the end.
     128    /// α *f(x) + β, α = 1.0,  β = 0.0 </param>
     129    /// <returns> An array containing the numeric constants.</returns>
     130    public static double[] ExtractConstants(ISymbolicExpressionTree tree, bool addLinearScalingConstants) {
     131      if (tree == null) throw new ArgumentNullException("tree");
     132      return ExtractConstants(tree.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>(), addLinearScalingConstants);
     133    }
     134
     135    /// <summary>
     136    /// Extracts all numeric constants from a list of nodes.
     137    /// </summary>
     138    /// <param name="nodes">The list of nodes for which the numeric constants should be extracted.</param>
     139    /// <param name="addLinearScalingConstants">Flag to determine whether constants for linear scaling have to be added at the end.
     140    /// α *f(x) + β, α = 1.0,  β = 0.0 </param>
     141    /// <returns> An array containing the numeric constants.</returns>
     142    public static double[] ExtractConstants(IEnumerable<ISymbolicExpressionTreeNode> nodes, bool addLinearScalingConstants) {
     143      if (nodes == null) throw new ArgumentNullException("nodes");
     144
    80145      var constants = new List<double>();
    81       foreach (var node in tree.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>()) {
     146      foreach (var node in nodes) {
    82147        ConstantTreeNode constantTreeNode = node as ConstantTreeNode;
    83148        VariableTreeNodeBase variableTreeNodeBase = node as VariableTreeNodeBase;
    … …  
    90155          for (int j = 0; j < factorVarTreeNode.Weights.Length; j++)
    91156            constants.Add(factorVarTreeNode.Weights[j]);
    92         } else throw new NotSupportedException(string.Format("Terminal nodes of type {0} are not supported.", node.GetType().GetPrettyName()));
    93       }
     157        } else throw new NotSupportedException(string.Format("Nodes of type {0} are not supported.", node.GetType().GetPrettyName()));
     158      }
     159      constants.Add(1.0);
     160      constants.Add(0.0);
    94161      return constants.ToArray();
    95162    }
    96163
    97     public static void UpdateConstants(ISymbolicExpressionTree tree, double[] constants) {
     164    /// <summary>
     165    /// Sets the numeric constants of the nodes to the provided values.
     166    /// </summary>
     167    /// <param name="nodes">The nodes whose constants should be updated.</param>
     168    /// <param name="constants">The numeric constants which should be set. </param>
     169    public static void UpdateConstants(IEnumerable<ISymbolicExpressionTreeNode> nodes, double[] constants) {
     170      if (nodes == null) throw new ArgumentNullException("nodes");
     171      if (constants == null) throw new ArgumentNullException("constants");
     172
    98173      int i = 0;
    99       foreach (var node in tree.Root.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>()) {
     174      foreach (var node in nodes) {
    100175        ConstantTreeNode constantTreeNode = node as ConstantTreeNode;
    101176        VariableTreeNodeBase variableTreeNodeBase = node as VariableTreeNodeBase;
    … …  
    111186      }
    112187    }
     188
     189    /// <summary>
     190    /// Sets all numeric constants of the symbolic expression tree to the provided values.
     191    /// </summary>
     192    /// <param name="tree">The tree for which the numeric constants should be updated.</param>
     193    /// <param name="constants">The numeric constants which should be set.</param>
     194    public static void UpdateConstants(ISymbolicExpressionTree tree, double[] constants) {
     195      if (tree == null) throw new ArgumentNullException("tree");
     196      if (constants == null) throw new ArgumentNullException("constants");
     197      UpdateConstants(tree.IterateNodesPrefix().OfType<SymbolicExpressionTreeTerminalNode>(), constants);
     198    }
    113199  }
    114200}

  • branches/2974_Constants_Optimization/HeuristicLab.Problems.DataAnalysis.Symbolic/3.4/HeuristicLab.Problems.DataAnalysis.Symbolic-3.4.csproj

    r16500 r16507  
    213213      <SubType>Code</SubType>
    214214    </Compile>
    215     <Compile Include="Constants Optimization\IConstantsOptimizer.cs" />
    216     <Compile Include="Constants Optimization\LMConstantsOptimizer.cs" />
    217     <Compile Include="Constants Optimization\Util.cs" />
     215    <Compile Include="ConstantsOptimization\IConstantsOptimizer.cs" />
     216    <Compile Include="ConstantsOptimization\LMConstantsOptimizer.cs" />
     217    <Compile Include="ConstantsOptimization\AutoDiffConverter.cs" />
     218    <Compile Include="ConstantsOptimization\Util.cs" />
     219    <Compile Include="ConstantsOptimization\VariableData.cs" />
    218220    <Compile Include="Converters\LinearModelToTreeConverter.cs" />
    219221    <Compile Include="Converters\TreeSimplifier.cs" />

  • branches/2974_Constants_Optimization/UnitTests/ConstantsOptimizationTests.cs

    r16500 r16507  
    3737
    3838    public static void CompareConstantsOptimizationResults(IRegressionProblemData problemData, ISymbolicExpressionTree tree) {
     39      var applyLinearScaling = true;
    3940      var old_optimizedTree = (ISymbolicExpressionTree)tree.Clone();
    4041      var old_result = SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(
    4142        new SymbolicDataAnalysisExpressionTreeLinearInterpreter(),
    42         old_optimizedTree, problemData, problemData.TrainingIndices, applyLinearScaling: true, maxIterations: 10);
     43        old_optimizedTree, problemData, problemData.TrainingIndices, applyLinearScaling, maxIterations: 10);
    4344
    4445
    4546      var new_optimizedTree = (ISymbolicExpressionTree)tree.Clone();
    46       var new_result = LMConstantsOptimizer.OptimizeConstants(new_optimizedTree, problemData, problemData.TrainingIndices, applyLinearScaling: true, maxIterations: 10);
     47      var new_result = LMConstantsOptimizer.OptimizeConstants(new_optimizedTree, problemData.Dataset, problemData.TargetVariable, problemData.TrainingIndices, applyLinearScaling, maxIterations: 10);
    4748
    4849      //check R² values
    … …  
    5051
    5152      //check numeric values of constants
    52       var old_constants = Util.ExtractConstants(old_optimizedTree);
    53       var new_constants = Util.ExtractConstants(new_optimizedTree);
     53      var old_constants = Util.ExtractConstants(old_optimizedTree, applyLinearScaling);
     54      var new_constants = Util.ExtractConstants(new_optimizedTree, applyLinearScaling);
    5455      //Assert.IsTrue(old_constants.SequenceEqual(new_constants));
    5556

  • branches/2974_Constants_Optimization/UnitTests/PerformanceTest.cs

    r16500 r16507  
    66using HeuristicLab.Problems.DataAnalysis;
    77using HeuristicLab.Problems.DataAnalysis.Symbolic;
     8using HeuristicLab.Problems.DataAnalysis.Symbolic.ConstantsOptimization;
    89using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
    910using HeuristicLab.Problems.Instances.DataAnalysis;
    … …  
    2324    [TestCategory("Problems.DataAnalysis.Symbolic.Regression")]
    2425    [TestProperty("Time", "long")]
    25     public void New_ConstantsOptimization_Tower_Algorithm() {
     26    public static void New_ConstantsOptimization_Tower_Algorithm() {
    2627      var twister = new MersenneTwister((uint)seed);
    2728      var problemData = new RegressionRealWorldInstanceProvider().LoadData(new Tower());
    … …  
    4041      //warm up
    4142      for (int i = 0; i < trees.Length; i++) {
    42         double quality = LMConstantsOptimizer.OptimizeConstants(trees[i], problemData, rows, true, maxIterations);
     43        if (!trees[i].IterateNodesPrefix().OfType<VariableTreeNode>().Any()) Debugger.Break();
     44        double quality = LMConstantsOptimizer.OptimizeConstants(trees[i], problemData.Dataset,problemData.TargetVariable, rows, true, maxIterations);
    4345      }
    4446
    … …  
    4749        watch.Start();
    4850        for (int i = 0; i < trees.Length; i++) {
    49           double quality = LMConstantsOptimizer.OptimizeConstants(trees[i], problemData, rows, true, maxIterations);
     51          double quality = LMConstantsOptimizer.OptimizeConstants(trees[i], problemData.Dataset, problemData.TargetVariable, rows, true, maxIterations);
    5052        }
    5153        watch.Stop();
    … …  
    7678      //warm up
    7779      for (int i = 0; i < trees.Length; i++) {
     80        if (!trees[i].IterateNodesPrefix().OfType<VariableTreeNode>().Any()) Debugger.Break();
    7881        double quality = SymbolicRegressionConstantOptimizationEvaluator.OptimizeConstants(
    7982          interpreter, trees[i], problemData, rows, true, maxIterations);

  • branches/2974_Constants_Optimization/UnitTests/UnitTests.csproj

    r16461 r16507  
    1111    <RootNamespace>UnitTests</RootNamespace>
    1212    <AssemblyName>UnitTests</AssemblyName>
    13     <TargetFrameworkVersion>v4.6.1</TargetFrameworkVersion>
     13    <TargetFrameworkVersion>v4.5</TargetFrameworkVersion>
    1414    <FileAlignment>512</FileAlignment>
    1515    <ProjectTypeGuids>{3AC096D0-A1C2-E12C-1390-A8335801FDAB};{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}</ProjectTypeGuids>
    … …  
    2121    <NuGetPackageImportStamp>
    2222    </NuGetPackageImportStamp>
     23    <TargetFrameworkProfile />
    2324  </PropertyGroup>
    2425  <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Debug|AnyCPU' ">
Note: See TracChangeset for help on using the changeset viewer.