Changeset 17074 for stable/HeuristicLab.Algorithms.DataAnalysis/3.4/Linear

Timestamp:

07/04/19 16:11:29 (5 years ago)

Author:

abeham

Message:

#2892: merged to stable

Location:

stable

Files:

• . (modified) (1 prop)
• HeuristicLab.Algorithms.DataAnalysis (modified) (1 prop)
• HeuristicLab.Algorithms.DataAnalysis/3.4 (modified) (1 prop)
• HeuristicLab.Algorithms.DataAnalysis/3.4/Linear/LinearRegression.cs (modified) (5 diffs)
• HeuristicLab.Algorithms.DataAnalysis/3.4/Linear/LinearRegressionModel.cs (copied) (copied from trunk/HeuristicLab.Algorithms.DataAnalysis/3.4/Linear/LinearRegressionModel.cs) (8 diffs)

stable/HeuristicLab.Algorithms.DataAnalysis/3.4/Linear/LinearRegression.cs

@@ -41 +41 @@
   [StorableClass]
   public sealed class LinearRegression : FixedDataAnalysisAlgorithm<IRegressionProblem> {
-    private const string LinearRegressionModelResultName = "Linear regression solution";
+    private const string SolutionResultName = "Linear regression solution";
+    private const string ConfidenceSolutionResultName = "Solution with prediction intervals";
 
     [StorableConstructor]
@@ -62 +63 @@
     protected override void Run(CancellationToken cancellationToken) {
       double rmsError, cvRmsError;
-      var solution = CreateLinearRegressionSolution(Problem.ProblemData, out rmsError, out cvRmsError);
-      Results.Add(new Result(LinearRegressionModelResultName, "The linear regression solution.", solution));
+      // produce both solutions, to allow symbolic manipulation of LR solutions as well
+      // as the calculation of prediction intervals.
+      // There is no clean way to implement the new model class for LR as a symbolic model.
+      var solution = CreateSolution(Problem.ProblemData, out rmsError, out cvRmsError);
+#pragma warning disable 168, 3021
+      var symbolicSolution = CreateLinearRegressionSolution(Problem.ProblemData, out rmsError, out cvRmsError);
+#pragma warning restore 168, 3021
+      Results.Add(new Result(SolutionResultName, "The linear regression solution.", symbolicSolution));
+      Results.Add(new Result(ConfidenceSolutionResultName, "Linear regression solution with parameter covariance matrix " +
+                                                           "and calculation of prediction intervals", solution));
       Results.Add(new Result("Root mean square error", "The root of the mean of squared errors of the linear regression solution on the training set.", new DoubleValue(rmsError)));
       Results.Add(new Result("Estimated root mean square error (cross-validation)", "The estimated root of the mean of squared errors of the linear regression solution via cross validation.", new DoubleValue(cvRmsError)));
     }
 
+    [Obsolete("Use CreateSolution() instead")]
     public static ISymbolicRegressionSolution CreateLinearRegressionSolution(IRegressionProblemData problemData, out double rmsError, out double cvRmsError) {
-      var dataset = problemData.Dataset;
-      string targetVariable = problemData.TargetVariable;
-      IEnumerable<string> allowedInputVariables = problemData.AllowedInputVariables;
-      IEnumerable<int> rows = problemData.TrainingIndices;
-      var doubleVariables = allowedInputVariables.Where(dataset.VariableHasType<double>);
-      var factorVariableNames = allowedInputVariables.Where(dataset.VariableHasType<string>);
-      var factorVariables = dataset.GetFactorVariableValues(factorVariableNames, rows);
-      double[,] binaryMatrix = dataset.ToArray(factorVariables, rows);
-      double[,] doubleVarMatrix = dataset.ToArray(doubleVariables.Concat(new string[] { targetVariable }), rows);
-      var inputMatrix = binaryMatrix.HorzCat(doubleVarMatrix);
-
-      if (inputMatrix.ContainsNanOrInfinity())
-        throw new NotSupportedException("Linear regression does not support NaN or infinity values in the input dataset.");
+      IEnumerable<string> doubleVariables;
+      IEnumerable<KeyValuePair<string, IEnumerable<string>>> factorVariables;
+      double[,] inputMatrix;
+      PrepareData(problemData, out inputMatrix, out doubleVariables, out factorVariables);
 
       alglib.linearmodel lm = new alglib.linearmodel();
@@ -87 +88 @@
       int nRows = inputMatrix.GetLength(0);
       int nFeatures = inputMatrix.GetLength(1) - 1;
-      double[] coefficients = new double[nFeatures + 1]; // last coefficient is for the constant
 
       int retVal = 1;
@@ -95 +95 @@
       cvRmsError = ar.cvrmserror;
 
+      double[] coefficients = new double[nFeatures + 1]; // last coefficient is for the constant
       alglib.lrunpack(lm, out coefficients, out nFeatures);
 
-      int nFactorCoeff = binaryMatrix.GetLength(1);
+      int nFactorCoeff = factorVariables.Sum(kvp => kvp.Value.Count());
       int nVarCoeff = doubleVariables.Count();
       var tree = LinearModelToTreeConverter.CreateTree(factorVariables, coefficients.Take(nFactorCoeff).ToArray(),
@@ -108 +109 @@
       return solution;
     }
+
+    public static IRegressionSolution CreateSolution(IRegressionProblemData problemData, out double rmsError, out double cvRmsError) {
+      IEnumerable<string> doubleVariables;
+      IEnumerable<KeyValuePair<string, IEnumerable<string>>> factorVariables;
+      double[,] inputMatrix;
+      PrepareData(problemData, out inputMatrix, out doubleVariables, out factorVariables);
+
+      alglib.linearmodel lm = new alglib.linearmodel();
+      alglib.lrreport ar = new alglib.lrreport();
+      int nRows = inputMatrix.GetLength(0);
+      int nFeatures = inputMatrix.GetLength(1) - 1;
+
+      int retVal = 1;
+      alglib.lrbuild(inputMatrix, nRows, nFeatures, out retVal, out lm, out ar);
+      if (retVal != 1) throw new ArgumentException("Error in calculation of linear regression solution");
+      rmsError = ar.rmserror;
+      cvRmsError = ar.cvrmserror;
+
+      // get parameters of the model
+      double[] w;
+      int nVars;
+      alglib.lrunpack(lm, out w, out nVars);
+
+      // ar.c is the covariation matrix,  array[0..NVars,0..NVars].
+      // C[i, j] = Cov(A[i], A[j])
+
+      var solution = new LinearRegressionModel(w, ar.c, cvRmsError, problemData.TargetVariable, doubleVariables, factorVariables)
+        .CreateRegressionSolution((IRegressionProblemData)problemData.Clone());
+      solution.Name = "Linear Regression Solution";
+      return solution;
+    }
+
+    private static void PrepareData(IRegressionProblemData problemData,
+      out double[,] inputMatrix,
+      out IEnumerable<string> doubleVariables,
+      out IEnumerable<KeyValuePair<string, IEnumerable<string>>> factorVariables) {
+      var dataset = problemData.Dataset;
+      string targetVariable = problemData.TargetVariable;
+      IEnumerable<string> allowedInputVariables = problemData.AllowedInputVariables;
+      IEnumerable<int> rows = problemData.TrainingIndices;
+      doubleVariables = allowedInputVariables.Where(dataset.VariableHasType<double>);
+      var factorVariableNames = allowedInputVariables.Where(dataset.VariableHasType<string>);
+      factorVariables = dataset.GetFactorVariableValues(factorVariableNames, rows);
+      double[,] binaryMatrix = dataset.ToArray(factorVariables, rows);
+      double[,] doubleVarMatrix = dataset.ToArray(doubleVariables.Concat(new string[] { targetVariable }), rows);
+      inputMatrix = binaryMatrix.HorzCat(doubleVarMatrix);
+
+      if (inputMatrix.ContainsNanOrInfinity())
+        throw new NotSupportedException("Linear regression does not support NaN or infinity values in the input dataset.");
+    }
     #endregion
   }

stable/HeuristicLab.Algorithms.DataAnalysis/3.4/Linear/LinearRegressionModel.cs

@@ -22 +22 @@
 using System;
 using System.Collections.Generic;
+using System.Drawing;
 using System.Linq;
 using HeuristicLab.Common;
@@ -35 +36 @@
   [Item("Linear Regression Model", "Represents a linear regression model.")]
   public sealed class LinearRegressionModel : RegressionModel, IConfidenceRegressionModel {
+    public static new Image StaticItemImage {
+      get { return HeuristicLab.Common.Resources.VSImageLibrary.Function; }
+    }
 
     [Storable]
@@ -49 +53 @@
       get; private set;
     }
-    
+
     public override IEnumerable<string> VariablesUsedForPrediction {
-      get { return allowedInputVariables; }
+      get { return doubleVariables.Union(factorVariables.Select(f => f.Key)); }
     }
 
     [Storable]
-    private string[] allowedInputVariables;
+    private string[] doubleVariables;
     [Storable]
     private List<KeyValuePair<string, IEnumerable<string>>> factorVariables;
+
+    /// <summary>
+    /// Enumerable of variable names used by the model including one-hot-encoded of factor variables.
+    /// </summary>
+    public IEnumerable<string> ParameterNames {
+      get {
+        return factorVariables.SelectMany(kvp => kvp.Value.Select(factorVal => $"{kvp.Key}={factorVal}"))
+          .Concat(doubleVariables)
+          .Concat(new[] { "<const>" });
+      }
+    }
 
     [StorableConstructor]
@@ -69 +84 @@
       this.NoiseSigma = original.NoiseSigma;
 
-      allowedInputVariables = (string[])original.allowedInputVariables.Clone();
+      doubleVariables = (string[])original.doubleVariables.Clone();
       this.factorVariables = original.factorVariables.Select(kvp => new KeyValuePair<string, IEnumerable<string>>(kvp.Key, new List<string>(kvp.Value))).ToList();
     }
@@ -78 +93 @@
       this.W = new double[w.Length];
       Array.Copy(w, W, w.Length);
-      this.C = new double[covariance.GetLength(0),covariance.GetLength(1)];
+      this.C = new double[covariance.GetLength(0), covariance.GetLength(1)];
       Array.Copy(covariance, C, covariance.Length);
       this.NoiseSigma = noiseSigma;
-      this.allowedInputVariables = doubleInputVariables.ToArray();
+      this.doubleVariables = doubleInputVariables.ToArray();
+      // clone
       this.factorVariables = factorVariables.Select(kvp => new KeyValuePair<string, IEnumerable<string>>(kvp.Key, new List<string>(kvp.Value))).ToList();
     }
@@ -94 +110 @@
 
     public override IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows) {
-      double[,] inputData = dataset.ToArray(allowedInputVariables, rows);
+      double[,] inputData = dataset.ToArray(doubleVariables, rows);
       double[,] factorData = dataset.ToArray(factorVariables, rows);
 
@@ -113 +129 @@
 
     public IEnumerable<double> GetEstimatedVariances(IDataset dataset, IEnumerable<int> rows) {
-      double[,] inputData = dataset.ToArray(allowedInputVariables, rows);
+      double[,] inputData = dataset.ToArray(doubleVariables, rows);
       double[,] factorData = dataset.ToArray(factorVariables, rows);
 
@@ -122 +138 @@
 
       double[] d = new double[C.GetLength(0)];
-      
+
       for (int row = 0; row < n; row++) {
         for (int column = 0; column < columns; column++) {
-          d[column] = inputData[row,column];
+          d[column] = inputData[row, column];
         }
         d[columns] = 1;
 
         double var = 0.0;
-        for(int i=0;i<d.Length;i++) {
-          for(int j = 0;j<d.Length;j++) {
+        for (int i = 0; i < d.Length; i++) {
+          for (int j = 0; j < d.Length; j++) {
             var += d[i] * C[i, j] * d[j];
           }
         }
-        yield return var + NoiseSigma*NoiseSigma;
+        yield return var + NoiseSigma * NoiseSigma;
       }
     }
-
 
     public override IRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData) {