Changeset 15030 for branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis/3.4/Implementation

Timestamp:

06/08/17 17:02:13 (8 years ago)

Author:

jzenisek

Message:

#2719 merged with trunk (revision 15029)

Location:

branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis

Files:

• . (modified) (1 prop)
• 3.4/Implementation/DataAnalysisProblemData.cs (modified) (5 diffs)
• 3.4/Implementation/Regression/RegressionSolutionVariableImpactsCalculator.cs (modified) (6 diffs)
• 3.4/Implementation/Transformations/LinearTransformation.cs (modified) (2 diffs)
• 3.4/Implementation/Transformations/ShiftStandardDistributionTransformation.cs (modified) (2 diffs)
• 3.4/Implementation/Transformations/ShiftToRangeTransformation.cs (modified) (2 diffs)
• 3.4/Implementation/Transformations/Transformation.cs (modified) (1 diff)

branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/DataAnalysisProblemData.cs

@@ -41 +41 @@
 
     #region parameter properites
+    //mkommend: inserted parameter caching due to performance reasons
+    private IFixedValueParameter<Dataset> datasetParameter;
     public IFixedValueParameter<Dataset> DatasetParameter {
-      get { return (IFixedValueParameter<Dataset>)Parameters[DatasetParameterName]; }
-    }
+      get {
+        if (datasetParameter == null) datasetParameter = (IFixedValueParameter<Dataset>)Parameters[DatasetParameterName];
+        return datasetParameter;
+      }
+    }
+
+    private IFixedValueParameter<ReadOnlyCheckedItemList<StringValue>> inputVariablesParameter;
     public IFixedValueParameter<ReadOnlyCheckedItemList<StringValue>> InputVariablesParameter {
-      get { return (IFixedValueParameter<ReadOnlyCheckedItemList<StringValue>>)Parameters[InputVariablesParameterName]; }
-    }
+      get {
+        if (inputVariablesParameter == null) inputVariablesParameter = (IFixedValueParameter<ReadOnlyCheckedItemList<StringValue>>)Parameters[InputVariablesParameterName];
+        return inputVariablesParameter;
+      }
+    }
+
+    private IFixedValueParameter<IntRange> trainingPartitionParameter;
     public IFixedValueParameter<IntRange> TrainingPartitionParameter {
-      get { return (IFixedValueParameter<IntRange>)Parameters[TrainingPartitionParameterName]; }
-    }
+      get {
+        if (trainingPartitionParameter == null) trainingPartitionParameter = (IFixedValueParameter<IntRange>)Parameters[TrainingPartitionParameterName];
+        return trainingPartitionParameter;
+      }
+    }
+
+    private IFixedValueParameter<IntRange> testPartitionParameter;
     public IFixedValueParameter<IntRange> TestPartitionParameter {
-      get { return (IFixedValueParameter<IntRange>)Parameters[TestPartitionParameterName]; }
-    }
+      get {
+        if (testPartitionParameter == null) testPartitionParameter = (IFixedValueParameter<IntRange>)Parameters[TestPartitionParameterName];
+        return testPartitionParameter;
+      }
+    }
+
     public IFixedValueParameter<ReadOnlyItemList<ITransformation>> TransformationsParameter {
       get { return (IFixedValueParameter<ReadOnlyItemList<ITransformation>>)Parameters[TransformationsParameterName]; }
@@ -73 +94 @@
     }
 
+    public double[,] AllowedInputsTrainingValues {
+      get { return Dataset.ToArray(AllowedInputVariables, TrainingIndices); }
+    }
+
+    public double[,] AllowedInputsTestValues { get { return Dataset.ToArray(AllowedInputVariables, TestIndices); } }
     public IntRange TrainingPartition {
       get { return TrainingPartitionParameter.Value; }
@@ -102 +128 @@
     public virtual bool IsTrainingSample(int index) {
       return index >= 0 && index < Dataset.Rows &&
-        TrainingPartition.Start <= index && index < TrainingPartition.End &&
-        (index < TestPartition.Start || TestPartition.End <= index);
+             TrainingPartition.Start <= index && index < TrainingPartition.End &&
+             (index < TestPartition.Start || TestPartition.End <= index);
     }
 
@@ -131 +157 @@
     protected DataAnalysisProblemData(IDataset dataset, IEnumerable<string> allowedInputVariables, IEnumerable<ITransformation> transformations = null) {
       if (dataset == null) throw new ArgumentNullException("The dataset must not be null.");
-      if (allowedInputVariables == null) throw new ArgumentNullException("The allowedInputVariables must not be null.");
-
-      if (allowedInputVariables.Except(dataset.DoubleVariables).Any())
-        throw new ArgumentException("All allowed input variables must be present in the dataset and of type double.");
-
-      var inputVariables = new CheckedItemList<StringValue>(dataset.DoubleVariables.Select(x => new StringValue(x)));
+      if (allowedInputVariables == null) throw new ArgumentNullException("The allowed input variables must not be null.");
+
+      if (allowedInputVariables.Except(dataset.DoubleVariables).Except(dataset.StringVariables).Any())
+        throw new ArgumentException("All allowed input variables must be present in the dataset and of type double or string.");
+
+      var variables = dataset.VariableNames.Where(variable => dataset.VariableHasType<double>(variable) || dataset.VariableHasType<string>(variable));
+      var inputVariables = new CheckedItemList<StringValue>(variables.Select(x => new StringValue(x)));
       foreach (StringValue x in inputVariables)
         inputVariables.SetItemCheckedState(x, allowedInputVariables.Contains(x.Value));
@@ -214 +241 @@
         InputVariables.SetItemCheckedState(inputVariable, variable != null && data.InputVariables.ItemChecked(variable));
       }
-
-      TrainingPartition.Start = TrainingPartition.End = 0;
-      TestPartition.Start = 0;
-      TestPartition.End = Dataset.Rows;
     }
   }

branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Regression/RegressionSolutionVariableImpactsCalculator.cs

@@ -42 +42 @@
       Noise
     }
-
+    public enum FactorReplacementMethodEnum {
+      Best,
+      Mode,
+      Shuffle
+    }
     public enum DataPartitionEnum {
       Training,
@@ -88 +92 @@
     }
 
-    public static IEnumerable<Tuple<string, double>> CalculateImpacts(IRegressionSolution solution,
+    public static IEnumerable<Tuple<string, double>> CalculateImpacts(
+      IRegressionSolution solution,
       DataPartitionEnum data = DataPartitionEnum.Training,
-      ReplacementMethodEnum replacement = ReplacementMethodEnum.Median) {
+      ReplacementMethodEnum replacementMethod = ReplacementMethodEnum.Median,
+      FactorReplacementMethodEnum factorReplacementMethod = FactorReplacementMethodEnum.Best) {
 
       var problemData = solution.ProblemData;
@@ -128 +134 @@
       var allowedInputVariables = dataset.VariableNames.Where(v => inputvariables.Contains(v)).ToList();
 
-      foreach (var inputVariable in allowedInputVariables) {
-        var newEstimates = EvaluateModelWithReplacedVariable(solution.Model, inputVariable, modifiableDataset, rows, replacement);
+      // calculate impacts for double variables
+      foreach (var inputVariable in allowedInputVariables.Where(problemData.Dataset.VariableHasType<double>)) {
+        var newEstimates = EvaluateModelWithReplacedVariable(solution.Model, inputVariable, modifiableDataset, rows, replacementMethod);
         var newR2 = OnlinePearsonsRCalculator.Calculate(targetValues, newEstimates, out error);
         if (error != OnlineCalculatorError.None) throw new InvalidOperationException("Error during R² calculation with replaced inputs.");
@@ -137 +144 @@
         impacts[inputVariable] = impact;
       }
+
+      // calculate impacts for string variables
+      foreach (var inputVariable in allowedInputVariables.Where(problemData.Dataset.VariableHasType<string>)) {
+        if (factorReplacementMethod == FactorReplacementMethodEnum.Best) {
+          // try replacing with all possible values and find the best replacement value
+          var smallestImpact = double.PositiveInfinity;
+          foreach (var repl in problemData.Dataset.GetStringValues(inputVariable, rows).Distinct()) {
+            var newEstimates = EvaluateModelWithReplacedVariable(solution.Model, inputVariable, modifiableDataset, rows,
+              Enumerable.Repeat(repl, dataset.Rows));
+            var newR2 = OnlinePearsonsRCalculator.Calculate(targetValues, newEstimates, out error);
+            if (error != OnlineCalculatorError.None)
+              throw new InvalidOperationException("Error during R² calculation with replaced inputs.");
+
+            newR2 = newR2 * newR2;
+            var impact = originalR2 - newR2;
+            if (impact < smallestImpact) smallestImpact = impact;
+          }
+          impacts[inputVariable] = smallestImpact;
+        } else {
+          // for replacement methods shuffle and mode
+          // calculate impacts for factor variables
+
+          var newEstimates = EvaluateModelWithReplacedVariable(solution.Model, inputVariable, modifiableDataset, rows,
+            factorReplacementMethod);
+          var newR2 = OnlinePearsonsRCalculator.Calculate(targetValues, newEstimates, out error);
+          if (error != OnlineCalculatorError.None)
+            throw new InvalidOperationException("Error during R² calculation with replaced inputs.");
+
+          newR2 = newR2 * newR2;
+          var impact = originalR2 - newR2;
+          impacts[inputVariable] = impact;
+        }
+      } // foreach
       return impacts.OrderByDescending(i => i.Value).Select(i => Tuple.Create(i.Key, i.Value));
     }
@@ -184 +224 @@
       }
 
-      dataset.ReplaceVariable(variable, replacementValues);
+      return EvaluateModelWithReplacedVariable(model, variable, dataset, rows, replacementValues);
+    }
+
+    private static IEnumerable<double> EvaluateModelWithReplacedVariable(
+      IRegressionModel model, string variable, ModifiableDataset dataset,
+      IEnumerable<int> rows,
+      FactorReplacementMethodEnum replacement = FactorReplacementMethodEnum.Shuffle) {
+      var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
+      List<string> replacementValues;
+      IRandom rand;
+
+      switch (replacement) {
+        case FactorReplacementMethodEnum.Mode:
+          var mostCommonValue = rows.Select(r => originalValues[r])
+            .GroupBy(v => v)
+            .OrderByDescending(g => g.Count())
+            .First().Key;
+          replacementValues = Enumerable.Repeat(mostCommonValue, dataset.Rows).ToList();
+          break;
+        case FactorReplacementMethodEnum.Shuffle:
+          // new var has same empirical distribution but the relation to y is broken
+          rand = new FastRandom(31415);
+          // prepare a complete column for the dataset
+          replacementValues = Enumerable.Repeat(string.Empty, dataset.Rows).ToList();
+          // shuffle only the selected rows
+          var shuffledValues = rows.Select(r => originalValues[r]).Shuffle(rand).ToList();
+          int i = 0;
+          // update column values 
+          foreach (var r in rows) {
+            replacementValues[r] = shuffledValues[i++];
+          }
+          break;
+        default:
+          throw new ArgumentException(string.Format("FactorReplacementMethod {0} cannot be handled.", replacement));
+      }
+
+      return EvaluateModelWithReplacedVariable(model, variable, dataset, rows, replacementValues);
+    }
+
+    private static IEnumerable<double> EvaluateModelWithReplacedVariable(IRegressionModel model, string variable,
+      ModifiableDataset dataset, IEnumerable<int> rows, IEnumerable<double> replacementValues) {
+      var originalValues = dataset.GetReadOnlyDoubleValues(variable).ToList();
+      dataset.ReplaceVariable(variable, replacementValues.ToList());
       //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
       var estimates = model.GetEstimatedValues(dataset, rows).ToList();
@@ -191 +273 @@
       return estimates;
     }
+    private static IEnumerable<double> EvaluateModelWithReplacedVariable(IRegressionModel model, string variable,
+      ModifiableDataset dataset, IEnumerable<int> rows, IEnumerable<string> replacementValues) {
+      var originalValues = dataset.GetReadOnlyStringValues(variable).ToList();
+      dataset.ReplaceVariable(variable, replacementValues.ToList());
+      //mkommend: ToList is used on purpose to avoid lazy evaluation that could result in wrong estimates due to variable replacements
+      var estimates = model.GetEstimatedValues(dataset, rows).ToList();
+      dataset.ReplaceVariable(variable, originalValues);
+
+      return estimates;
+    }
   }
 }

branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Transformations/LinearTransformation.cs

@@ -52 +52 @@
     public double Multiplier {
       get { return MultiplierParameter.Value.Value; }
-      protected set {
+      set {
         MultiplierParameter.Value.Value = value;
       }
@@ -59 +59 @@
     public double Addend {
       get { return AddendParameter.Value.Value; }
-      protected set {
+      set {
         AddendParameter.Value.Value = value;
       }

branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Transformations/ShiftStandardDistributionTransformation.cs

@@ -71 +71 @@
 
     public override IEnumerable<double> Apply(IEnumerable<double> data) {
-      ConfigureParameters(data);
       if (OriginalStandardDeviation.IsAlmost(0.0)) {
         return data;
@@ -94 +93 @@
     }
 
-    protected void ConfigureParameters(IEnumerable<double> data) {
+    public override void ConfigureParameters(IEnumerable<double> data) {
       OriginalStandardDeviation = data.StandardDeviation();
       OriginalMean = data.Average();

branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Transformations/ShiftToRangeTransformation.cs

@@ -44 +44 @@
     }
 
-    public override IEnumerable<double> Apply(IEnumerable<double> data) {
-      ConfigureParameters(data);
-      return base.Apply(data);
-    }
-
     public override bool Check(IEnumerable<double> data, out string errorMsg) {
       ConfigureParameters(data);
@@ -54 +49 @@
     }
 
-    protected void ConfigureParameters(IEnumerable<double> data) {
+    public override void ConfigureParameters(IEnumerable<double> data) {
       double originalRangeStart = data.Min();
       double originalRangeEnd = data.Max();

branches/HeuristicLab.DatastreamAnalysis/HeuristicLab.Problems.DataAnalysis/3.4/Implementation/Transformations/Transformation.cs

@@ -66 +66 @@
     protected Transformation(IEnumerable<string> allowedColumns) : base(allowedColumns) { }
 
+    public virtual void ConfigureParameters(IEnumerable<T> data) {
+      // override in transformations with parameters
+    }
+
     public abstract IEnumerable<T> Apply(IEnumerable<T> data);
+    public IEnumerable<T> ConfigureAndApply(IEnumerable<T> data) {
+      ConfigureParameters(data);
+      return Apply(data);
+    }
 
     public abstract bool Check(IEnumerable<T> data, out string errorMsg);