source: stable/HeuristicLab.Algorithms.DataAnalysis/3.4/Linear/LinearRegression.cs @ 17074

Last change on this file since 17074 was 17074, checked in by abeham, 23 months ago

#2892: merged to stable

File size: 8.1 KB
RevLine 
[5617]1#region License Information
2/* HeuristicLab
[15584]3 * Copyright (C) 2002-2018 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
[5617]4 *
5 * This file is part of HeuristicLab.
6 *
7 * HeuristicLab is free software: you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation, either version 3 of the License, or
10 * (at your option) any later version.
11 *
12 * HeuristicLab is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with HeuristicLab. If not, see <http://www.gnu.org/licenses/>.
19 */
20#endregion
21
22using System;
[5777]23using System.Collections.Generic;
[5617]24using System.Linq;
[15061]25using System.Threading;
[5617]26using HeuristicLab.Common;
27using HeuristicLab.Core;
28using HeuristicLab.Data;
29using HeuristicLab.Optimization;
30using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
31using HeuristicLab.Problems.DataAnalysis;
32using HeuristicLab.Problems.DataAnalysis.Symbolic;
[5624]33using HeuristicLab.Problems.DataAnalysis.Symbolic.Regression;
[5617]34
35namespace HeuristicLab.Algorithms.DataAnalysis {
36  /// <summary>
37  /// Linear regression data analysis algorithm.
38  /// </summary>
[13297]39  [Item("Linear Regression (LR)", "Linear regression data analysis algorithm (wrapper for ALGLIB).")]
[12708]40  [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 100)]
[5617]41  [StorableClass]
42  public sealed class LinearRegression : FixedDataAnalysisAlgorithm<IRegressionProblem> {
[17074]43    private const string SolutionResultName = "Linear regression solution";
44    private const string ConfidenceSolutionResultName = "Solution with prediction intervals";
[5617]45
46    [StorableConstructor]
47    private LinearRegression(bool deserializing) : base(deserializing) { }
48    private LinearRegression(LinearRegression original, Cloner cloner)
49      : base(original, cloner) {
50    }
51    public LinearRegression()
52      : base() {
[5649]53      Problem = new RegressionProblem();
[5617]54    }
55    [StorableHook(HookType.AfterDeserialization)]
56    private void AfterDeserialization() { }
57
58    public override IDeepCloneable Clone(Cloner cloner) {
59      return new LinearRegression(this, cloner);
60    }
61
62    #region linear regression
[15061]63    protected override void Run(CancellationToken cancellationToken) {
[5617]64      double rmsError, cvRmsError;
[17074]65      // produce both solutions, to allow symbolic manipulation of LR solutions as well
66      // as the calculation of prediction intervals.
67      // There is no clean way to implement the new model class for LR as a symbolic model.
68      var solution = CreateSolution(Problem.ProblemData, out rmsError, out cvRmsError);
69#pragma warning disable 168, 3021
70      var symbolicSolution = CreateLinearRegressionSolution(Problem.ProblemData, out rmsError, out cvRmsError);
71#pragma warning restore 168, 3021
72      Results.Add(new Result(SolutionResultName, "The linear regression solution.", symbolicSolution));
73      Results.Add(new Result(ConfidenceSolutionResultName, "Linear regression solution with parameter covariance matrix " +
74                                                           "and calculation of prediction intervals", solution));
[5649]75      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)));
76      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)));
[5617]77    }
78
[17074]79    [Obsolete("Use CreateSolution() instead")]
[5624]80    public static ISymbolicRegressionSolution CreateLinearRegressionSolution(IRegressionProblemData problemData, out double rmsError, out double cvRmsError) {
[17074]81      IEnumerable<string> doubleVariables;
82      IEnumerable<KeyValuePair<string, IEnumerable<string>>> factorVariables;
83      double[,] inputMatrix;
84      PrepareData(problemData, out inputMatrix, out doubleVariables, out factorVariables);
[15131]85
[5617]86      alglib.linearmodel lm = new alglib.linearmodel();
87      alglib.lrreport ar = new alglib.lrreport();
88      int nRows = inputMatrix.GetLength(0);
89      int nFeatures = inputMatrix.GetLength(1) - 1;
90
91      int retVal = 1;
92      alglib.lrbuild(inputMatrix, nRows, nFeatures, out retVal, out lm, out ar);
[5649]93      if (retVal != 1) throw new ArgumentException("Error in calculation of linear regression solution");
[5617]94      rmsError = ar.rmserror;
95      cvRmsError = ar.cvrmserror;
96
[17074]97      double[] coefficients = new double[nFeatures + 1]; // last coefficient is for the constant
[5617]98      alglib.lrunpack(lm, out coefficients, out nFeatures);
99
[17074]100      int nFactorCoeff = factorVariables.Sum(kvp => kvp.Value.Count());
[15142]101      int nVarCoeff = doubleVariables.Count();
102      var tree = LinearModelToTreeConverter.CreateTree(factorVariables, coefficients.Take(nFactorCoeff).ToArray(),
[15788]103        doubleVariables.ToArray(), coefficients.Skip(nFactorCoeff).Take(nVarCoeff).ToArray(),
[15142]104        @const: coefficients[nFeatures]);
[15788]105
[14795]106      SymbolicRegressionSolution solution = new SymbolicRegressionSolution(new SymbolicRegressionModel(problemData.TargetVariable, tree, new SymbolicDataAnalysisExpressionTreeLinearInterpreter()), (IRegressionProblemData)problemData.Clone());
[6555]107      solution.Model.Name = "Linear Regression Model";
[7588]108      solution.Name = "Linear Regression Solution";
[5624]109      return solution;
[5617]110    }
[17074]111
112    public static IRegressionSolution CreateSolution(IRegressionProblemData problemData, out double rmsError, out double cvRmsError) {
113      IEnumerable<string> doubleVariables;
114      IEnumerable<KeyValuePair<string, IEnumerable<string>>> factorVariables;
115      double[,] inputMatrix;
116      PrepareData(problemData, out inputMatrix, out doubleVariables, out factorVariables);
117
118      alglib.linearmodel lm = new alglib.linearmodel();
119      alglib.lrreport ar = new alglib.lrreport();
120      int nRows = inputMatrix.GetLength(0);
121      int nFeatures = inputMatrix.GetLength(1) - 1;
122
123      int retVal = 1;
124      alglib.lrbuild(inputMatrix, nRows, nFeatures, out retVal, out lm, out ar);
125      if (retVal != 1) throw new ArgumentException("Error in calculation of linear regression solution");
126      rmsError = ar.rmserror;
127      cvRmsError = ar.cvrmserror;
128
129      // get parameters of the model
130      double[] w;
131      int nVars;
132      alglib.lrunpack(lm, out w, out nVars);
133
134      // ar.c is the covariation matrix,  array[0..NVars,0..NVars].
135      // C[i, j] = Cov(A[i], A[j])
136
137      var solution = new LinearRegressionModel(w, ar.c, cvRmsError, problemData.TargetVariable, doubleVariables, factorVariables)
138        .CreateRegressionSolution((IRegressionProblemData)problemData.Clone());
139      solution.Name = "Linear Regression Solution";
140      return solution;
141    }
142
143    private static void PrepareData(IRegressionProblemData problemData,
144      out double[,] inputMatrix,
145      out IEnumerable<string> doubleVariables,
146      out IEnumerable<KeyValuePair<string, IEnumerable<string>>> factorVariables) {
147      var dataset = problemData.Dataset;
148      string targetVariable = problemData.TargetVariable;
149      IEnumerable<string> allowedInputVariables = problemData.AllowedInputVariables;
150      IEnumerable<int> rows = problemData.TrainingIndices;
151      doubleVariables = allowedInputVariables.Where(dataset.VariableHasType<double>);
152      var factorVariableNames = allowedInputVariables.Where(dataset.VariableHasType<string>);
153      factorVariables = dataset.GetFactorVariableValues(factorVariableNames, rows);
154      double[,] binaryMatrix = dataset.ToArray(factorVariables, rows);
155      double[,] doubleVarMatrix = dataset.ToArray(doubleVariables.Concat(new string[] { targetVariable }), rows);
156      inputMatrix = binaryMatrix.HorzCat(doubleVarMatrix);
157
158      if (inputMatrix.ContainsNanOrInfinity())
159        throw new NotSupportedException("Linear regression does not support NaN or infinity values in the input dataset.");
160    }
[5617]161    #endregion
162  }
163}
Note: See TracBrowser for help on using the repository browser.