Free cookie consent management tool by TermsFeed Policy Generator

Ignore:
Timestamp:
02/20/14 14:47:17 (10 years ago)
Author:
gkronber
Message:

#2093 added a notice that the log residual evaluator does not apply linear scaling

File:
1 edited

Legend:

Unmodified
Added
Removed
  • branches/LogResidualEvaluator/HeuristicLab.Problems.DataAnalysis.Symbolic.Regression/3.4/SingleObjective/Evaluators/SymbolicRegressionLogResidualEvaluator.cs

    r10305 r10482  
    3030
    3131namespace HeuristicLab.Problems.DataAnalysis.Symbolic.Regression {
    32   [Item("Log Residual Evaluator", "Evaluator for symbolic regression models that calculates the mean of logarithmic absolute residuals avg(log( 1 + abs(y' - y)))." +
     32  [Item("Log Residual Evaluator", "Evaluator for symbolic regression models that calculates the mean of logarithmic absolute residuals avg(log( 1 + abs(y' - y)))" +
     33                                  "This evaluator does not perform linear scaling!" +
    3334                                  "This evaluator can be useful if the modeled function contains discontinuities (e.g. 1/x). " +
    34                                   "In preliminary experiments, we found that this improves success rates for data sets with a few very large/small target values " +
    35                                   "(e.g. Korns benchmark instances containing inverses of near zero values). In these cases the squared error or absolute " +
    36                                   "error puts too much emphasis on modeling the outlier values. Using log-residuals instead of e.g. squared residuals has the " +
    37                                   "effect that smaller residuals have a stronger impact on the total quality compared to the very large residuals occuring with such extreme values." +
     35                                  "For some data sets (e.g. Korns benchmark instances containing inverses of near zero values) the squared error or absolute " +
     36                                  "error put too much emphasis on modeling the outlier values. Using log-residuals instead has the " +
     37                                  "effect that smaller residuals have a stronger impact on the total quality compared to the large residuals." +
    3838                                  "This effects GP convergence because functional fragments which are necessary to explain small variations are also more likely" +
    3939                                  " to stay in the population. This is useful even when the actual objective function is mean of squared errors.")]
     
    5757      IEnumerable<int> rows = GenerateRowsToEvaluate();
    5858
    59       double quality = Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, ProblemDataParameter.ActualValue, rows, ApplyLinearScalingParameter.ActualValue.Value);
     59      double quality = Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, solution, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, ProblemDataParameter.ActualValue, rows);
    6060      QualityParameter.ActualValue = new DoubleValue(quality);
    6161
     
    6363    }
    6464
    65     public static double Calculate(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, ISymbolicExpressionTree solution, double lowerEstimationLimit, double upperEstimationLimit, IRegressionProblemData problemData, IEnumerable<int> rows, bool applyLinearScaling) {
     65    public static double Calculate(ISymbolicDataAnalysisExpressionTreeInterpreter interpreter, ISymbolicExpressionTree solution, double lowerEstimationLimit, double upperEstimationLimit, IRegressionProblemData problemData, IEnumerable<int> rows) {
    6666      IEnumerable<double> estimatedValues = interpreter.GetSymbolicExpressionTreeValues(solution, problemData.Dataset, rows);
    6767      IEnumerable<double> targetValues = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
     
    8282      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = context;
    8383      EstimationLimitsParameter.ExecutionContext = context;
    84       ApplyLinearScalingParameter.ExecutionContext = context;
    8584
    86       double mlr = Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, tree, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, problemData, rows, ApplyLinearScalingParameter.ActualValue.Value);
     85      double mlr = Calculate(SymbolicDataAnalysisTreeInterpreterParameter.ActualValue, tree, EstimationLimitsParameter.ActualValue.Lower, EstimationLimitsParameter.ActualValue.Upper, problemData, rows);
    8786
    8887      SymbolicDataAnalysisTreeInterpreterParameter.ExecutionContext = null;
    8988      EstimationLimitsParameter.ExecutionContext = null;
    90       ApplyLinearScalingParameter.ExecutionContext = null;
    9189
    9290      return mlr;
Note: See TracChangeset for help on using the changeset viewer.