Free cookie consent management tool by TermsFeed Policy Generator

source: branches/2789_MathNetNumerics-Exploration/HeuristicLab.Algorithms.DataAnalysis.Experimental/GAM.cs @ 17778

Last change on this file since 17778 was 15469, checked in by gkronber, 7 years ago

#2789 trying to get SBART to work correctly.
File size: 16.0 KB
Line 
1#region License Information
2/* HeuristicLab
3 * Copyright (C) 2002-2016 Heuristic and Evolutionary Algorithms Laboratory (HEAL)
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;
23using System.Collections.Generic;
24using System.Linq;
25using System.Threading;
26using HeuristicLab.Analysis;
27using HeuristicLab.Common;
28using HeuristicLab.Core;
29using HeuristicLab.Data;
30using HeuristicLab.Optimization;
31using HeuristicLab.Parameters;
32using HeuristicLab.Persistence.Default.CompositeSerializers.Storable;
33using HeuristicLab.Problems.DataAnalysis;
34
35namespace HeuristicLab.Algorithms.DataAnalysis.Experimental {
36  // UNFINISHED
37  [Item("Generalized Additive Modelling", "GAM")]
38  [Creatable(CreatableAttribute.Categories.DataAnalysisRegression, Priority = 102)]
39  [StorableClass]
40  public sealed class GAM : FixedDataAnalysisAlgorithm<IRegressionProblem> {
41
42    private const string LambdaParameterName = "Lambda";
43    private const string MaxIterationsParameterName = "Max iterations";
44    private const string MaxInteractionsParameterName = "Max interactions";
45
46    public IFixedValueParameter<DoubleValue> LambdaParameter {
47      get { return (IFixedValueParameter<DoubleValue>)Parameters[LambdaParameterName]; }
48    }
49    public IFixedValueParameter<IntValue> MaxIterationsParameter {
50      get { return (IFixedValueParameter<IntValue>)Parameters[MaxIterationsParameterName]; }
51    }
52    public IFixedValueParameter<IntValue> MaxInteractionsParameter {
53      get { return (IFixedValueParameter<IntValue>)Parameters[MaxInteractionsParameterName]; }
54    }
55
56    public double Lambda {
57      get { return LambdaParameter.Value.Value; }
58      set { LambdaParameter.Value.Value = value; }
59    }
60    public int MaxIterations {
61      get { return MaxIterationsParameter.Value.Value; }
62      set { MaxIterationsParameter.Value.Value = value; }
63    }
64    public int MaxInteractions {
65      get { return MaxInteractionsParameter.Value.Value; }
66      set { MaxInteractionsParameter.Value.Value = value; }
67    }
68
69    [StorableConstructor]
70    private GAM(bool deserializing) : base(deserializing) { }
71    [StorableHook(HookType.AfterDeserialization)]
72    private void AfterDeserialization() {
73    }
74
75    private GAM(GAM original, Cloner cloner)
76      : base(original, cloner) {
77    }
78    public override IDeepCloneable Clone(Cloner cloner) {
79      return new GAM(this, cloner);
80    }
81
82    public GAM()
83      : base() {
84      Problem = new RegressionProblem();
85      Parameters.Add(new FixedValueParameter<DoubleValue>(LambdaParameterName, "Regularization for smoothing splines", new DoubleValue(1.0)));
86      Parameters.Add(new FixedValueParameter<IntValue>(MaxIterationsParameterName, "", new IntValue(100)));
87      Parameters.Add(new FixedValueParameter<IntValue>(MaxInteractionsParameterName, "", new IntValue(1)));
88    }
89
90
91    protected override void Run(CancellationToken cancellationToken) {
92      double lambda = Lambda;
93      int maxIters = MaxIterations;
94      int maxInteractions = MaxInteractions;
95      if (maxInteractions < 1 || maxInteractions > 5) throw new ArgumentException("Max interactions is outside the valid range [1 .. 5]");
96
97      // calculates a GAM model using a linear representation + independent non-linear functions of each variable
98      // using backfitting algorithm (see The Elements of Statistical Learning page 298)
99
100      var problemData = Problem.ProblemData;
101      var y = problemData.TargetVariableTrainingValues.ToArray();
102      var avgY = y.Average();
103      var inputVars = Problem.ProblemData.AllowedInputVariables.ToArray();
104      var nTerms = 0; // inputVars.Length; // LR
105      for (int i = 1; i <= maxInteractions; i++) {
106        nTerms += inputVars.Combinations(i).Count();
107      }
108
109      IRegressionModel[] f = new IRegressionModel[nTerms];
110      for (int i = 0; i < f.Length; i++) {
111        f[i] = new ConstantModel(0.0, problemData.TargetVariable);
112      }
113
114      var rmseTable = new DataTable("RMSE");
115      var rmseRow = new DataRow("RMSE (train)");
116      var rmseRowTest = new DataRow("RMSE (test)");
117      rmseTable.Rows.Add(rmseRow);
118      rmseTable.Rows.Add(rmseRowTest);
119
120      Results.Add(new Result("RMSE", rmseTable));
121      rmseRow.Values.Add(CalculateResiduals(problemData, f, -1, avgY, problemData.TrainingIndices).StandardDeviation()); // -1 index to use all predictors
122      rmseRowTest.Values.Add(CalculateResiduals(problemData, f, -1, avgY, problemData.TestIndices).StandardDeviation());
123
124      // for analytics
125      double[] rss = new double[f.Length];
126      string[] terms = new string[f.Length];
127      Results.Add(new Result("RSS Values", typeof(DoubleMatrix)));
128
129      var combinations = new List<string[]>();
130      for (int i = 1; i <= maxInteractions; i++)
131        combinations.AddRange(HeuristicLab.Common.EnumerableExtensions.Combinations(inputVars, i).Select(c => c.ToArray()));
132      // combinations.Add(new string[] { "X1", "X2" });
133      // combinations.Add(new string[] { "X3", "X4" });
134      // combinations.Add(new string[] { "X5", "X6" });
135      // combinations.Add(new string[] { "X1", "X7", "X9" });
136      // combinations.Add(new string[] { "X3", "X6", "X10" });
137
138
139
140      // until convergence
141      int iters = 0;
142      var t = new double[y.Length];
143      while (iters++ < maxIters) {
144        int j = 0;
145        //foreach (var inputVar in inputVars) {
146        //  var res = CalculateResiduals(problemData, f, j, avgY, problemData.TrainingIndices);
147        //  rss[j] = res.Variance();
148        //  terms[j] = inputVar;
149        //  f[j] = RegressLR(problemData, inputVar, res);
150        //  j++;
151        //}
152
153
154
155        foreach (var element in combinations) {
156          var res = CalculateResiduals(problemData, f, j, avgY, problemData.TrainingIndices);
157          rss[j] = res.Variance();
158          terms[j] = string.Format("f({0})", string.Join(",", element));
159          f[j] = RegressSpline(problemData, element.ToArray(), res, lambda);
160          j++;
161        }
162
163        rmseRow.Values.Add(CalculateResiduals(problemData, f, -1, avgY, problemData.TrainingIndices).StandardDeviation()); // -1 index to use all predictors
164        rmseRowTest.Values.Add(CalculateResiduals(problemData, f, -1, avgY, problemData.TestIndices).StandardDeviation());
165
166        // calculate table with residual contributions of each term
167        var rssTable = new DoubleMatrix(rss.Length, 1, new string[] { "RSS" }, terms);
168        for (int i = 0; i < rss.Length; i++) rssTable[i, 0] = rss[i];
169        Results["RSS Values"].Value = rssTable;
170
171        if (cancellationToken.IsCancellationRequested) break;
172      }
173
174      var model = new RegressionEnsembleModel(f.Concat(new[] { new ConstantModel(avgY, problemData.TargetVariable) }));
175      model.AverageModelEstimates = false;
176      var solution = model.CreateRegressionSolution((IRegressionProblemData)problemData.Clone());
177      Results.Add(new Result("Ensemble solution", solution));
178    }
179
180    private double[] CalculateResiduals(IRegressionProblemData problemData, IRegressionModel[] f, int j, double avgY, IEnumerable<int> rows) {
181      var y = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
182      double[] t = y.Select(yi => yi - avgY).ToArray();
183      // collect other predictions
184      for (int k = 0; k < f.Length; k++) {
185        if (k != j) {
186          var pred = f[k].GetEstimatedValues(problemData.Dataset, rows).ToArray();
187          // determine target for this smoother
188          for (int i = 0; i < t.Length; i++) {
189            t[i] -= pred[i];
190          }
191        }
192      }
193      return t;
194    }
195
196    private IRegressionModel RegressLR(IRegressionProblemData problemData, string inputVar, double[] target) {
197      // Umständlich!
198      var ds = ((Dataset)problemData.Dataset).ToModifiable();
199      ds.ReplaceVariable(problemData.TargetVariable, target.Concat(Enumerable.Repeat(0.0, ds.Rows - target.Length)).ToList<double>());
200      var pd = new RegressionProblemData(ds, new string[] { inputVar }, problemData.TargetVariable);
201      pd.TrainingPartition.Start = problemData.TrainingPartition.Start;
202      pd.TrainingPartition.End = problemData.TrainingPartition.End;
203      pd.TestPartition.Start = problemData.TestPartition.Start;
204      pd.TestPartition.End = problemData.TestPartition.End;
205      double rmsError, cvRmsError;
206      return LinearRegression.CreateLinearRegressionSolution(pd, out rmsError, out cvRmsError).Model;
207    }
208
209    // private IRegressionModel RegressSpline(IRegressionProblemData problemData, string inputVar, double[] target, double lambda) {
210    //   if (problemData.Dataset.VariableHasType<double>(inputVar)) {
211    //     // Umständlich!
212    //     return Splines.CalculatePenalizedRegressionSpline(
213    //       problemData.Dataset.GetDoubleValues(inputVar, problemData.TrainingIndices).ToArray(),
214    //       (double[])target.Clone(), lambda,
215    //       problemData.TargetVariable, new string[] { inputVar }
216    //       );
217    //   } else return new ConstantModel(target.Average(), problemData.TargetVariable);
218    // }
219    private IRegressionModel RegressSpline(IRegressionProblemData problemData, string[] inputVars, double[] target, double lambda) {
220      if (inputVars.All(problemData.Dataset.VariableHasType<double>)) {
221        var product = problemData.Dataset.GetDoubleValues(inputVars.First(), problemData.TrainingIndices).ToArray();
222        for (int i = 1; i < inputVars.Length; i++) {
223          product = product.Zip(problemData.Dataset.GetDoubleValues(inputVars[i], problemData.TrainingIndices), (pi, vi) => pi * vi).ToArray();
224        }
225        // CubicSplineGCV.CubGcvReport report;
226        // return CubicSplineGCV.CalculateCubicSpline(
227        //   product,
228        //   (double[])target.Clone(),
229        //   problemData.TargetVariable, inputVars, out report
230        //   );
231        //
232        // double optTolerance; double cvRMSE;
233        // find tolerance
234        // var ensemble = Splines.CalculateSmoothingSplineReinsch(product, (double[])target.Clone(), inputVars, problemData.TargetVariable, out optTolerance, out cvRMSE);
235        // // train on whole data
236        // return Splines.CalculateSmoothingSplineReinsch(product, (double[])target.Clone(), inputVars, optTolerance, product.Length - 1, problemData.TargetVariable);
237
238
239        // find tolerance
240        //var bestLambda = double.NaN;
241        // double bestCVRMSE = target.StandardDeviation();
242        // double avgTrainRMSE = double.PositiveInfinity;
243        // double[] bestPredictions = new double[target.Length]; // zero
244
245
246        //double[] bestSSE = target.Select(ti => ti*ti).ToArray(); // target - zero
247        //for (double curLambda = 6.0; curLambda >= -6.0; curLambda -= 1.0) {
248        //  double[] predictions;
249        //  var ensemble = Splines.CalculatePenalizedRegressionSpline(product, (double[])target.Clone(), curLambda, problemData.TargetVariable, inputVars, out avgTrainRMSE, out cvRMSE, out predictions);
250        //  double[] sse = target.Zip(predictions, (t, p) => (t - p)*(t-p)).ToArray();
251        //  // Console.Write("{0} {1} {2}", curLambda, avgTrainRMSE, cvRMSE);
252        //  double bothTails = .0, leftTail = .0, rightTail = .0;
253        //  alglib.stest.onesamplesigntest(bestSSE.Zip(sse, (a, b) => a-b).ToArray(), predictions.Length, 0.0, ref bothTails, ref leftTail, ref rightTail);
254        //  if (bothTails < 0.1 && bestCVRMSE > cvRMSE) {
255        //    Console.Write(" *");
256        //    bestCVRMSE = cvRMSE;
257        //    bestLambda = curLambda;
258        //    bestSSE = sse;
259        //    bestPredictions = predictions;
260        //  }
261        //  // Console.WriteLine();
262        //}
263        //if (double.IsNaN(bestLambda)) {
264        //  return new ConstantModel(target.Average(), problemData.TargetVariable);
265        //} else {
266        // train on whole data
267
268
269        // return Splines.CalculatePenalizedRegressionSpline(product, (double[])target.Clone(), lambda, problemData.TargetVariable, inputVars, out avgTrainRMSE, out cvRMSE, out bestPredictions);
270        SBART.SBART_Report rep;
271        var w = product.Select(_ => 1.0).ToArray();
272        var model = SBART.CalculateSBART(product, (double[])target.Clone(), w, 10, problemData.TargetVariable, inputVars, out rep);
273        Console.WriteLine("{0} {1:N5} {2:N5} {3:N5} {4:N5}", string.Join(",", inputVars), rep.gcv, rep.leverage.Sum(), product.StandardDeviation(), target.StandardDeviation());
274        return model;
275        // }
276
277      } else return new ConstantModel(target.Average(), problemData.TargetVariable);
278    }
279
280    private IRegressionModel RegressRF(IRegressionProblemData problemData, string inputVar, double[] target, double lambda) {
281      if (problemData.Dataset.VariableHasType<double>(inputVar)) {
282        // Umständlich!
283        var ds = ((Dataset)problemData.Dataset).ToModifiable();
284        ds.ReplaceVariable(problemData.TargetVariable, target.Concat(Enumerable.Repeat(0.0, ds.Rows - target.Length)).ToList<double>());
285        var pd = new RegressionProblemData(ds, new string[] { inputVar }, problemData.TargetVariable);
286        pd.TrainingPartition.Start = problemData.TrainingPartition.Start;
287        pd.TrainingPartition.End = problemData.TrainingPartition.End;
288        pd.TestPartition.Start = problemData.TestPartition.Start;
289        pd.TestPartition.End = problemData.TestPartition.End;
290        double rmsError, oobRmsError;
291        double avgRelError, oobAvgRelError;
292        return RandomForestRegression.CreateRandomForestRegressionModel(pd, 100, 0.5, 0.5, 1234, out rmsError, out avgRelError, out oobRmsError, out oobAvgRelError);
293      } else return new ConstantModel(target.Average(), problemData.TargetVariable);
294    }
295  }
296
297
298  // UNFINISHED
299  public class RBFModel : NamedItem, IRegressionModel {
300    private alglib.rbfmodel model;
301
302    public string TargetVariable { get; set; }
303
304    public IEnumerable<string> VariablesUsedForPrediction { get; private set; }
305    private ITransformation<double>[] scaling;
306
307    public event EventHandler TargetVariableChanged;
308
309    public RBFModel(RBFModel orig, Cloner cloner) : base(orig, cloner) {
310      this.TargetVariable = orig.TargetVariable;
311      this.VariablesUsedForPrediction = orig.VariablesUsedForPrediction.ToArray();
312      this.model = (alglib.rbfmodel)orig.model.make_copy();
313      this.scaling = orig.scaling.Select(s => cloner.Clone(s)).ToArray();
314    }
315    public RBFModel(alglib.rbfmodel model, string targetVar, string[] inputs, IEnumerable<ITransformation<double>> scaling) : base("RBFModel", "RBFModel") {
316      this.model = model;
317      this.TargetVariable = targetVar;
318      this.VariablesUsedForPrediction = inputs;
319      this.scaling = scaling.ToArray();
320    }
321
322    public override IDeepCloneable Clone(Cloner cloner) {
323      return new RBFModel(this, cloner);
324    }
325
326    public IRegressionSolution CreateRegressionSolution(IRegressionProblemData problemData) {
327      return new RegressionSolution(this, (IRegressionProblemData)problemData.Clone());
328    }
329
330    public IEnumerable<double> GetEstimatedValues(IDataset dataset, IEnumerable<int> rows) {
331      double[] x = new double[VariablesUsedForPrediction.Count()];
332      double[] y;
333      foreach (var r in rows) {
334        int c = 0;
335        foreach (var v in VariablesUsedForPrediction) {
336          x[c] = scaling[c].Apply(dataset.GetDoubleValue(v, r).ToEnumerable()).First(); // OUCH!
337          c++;
338        }
339        alglib.rbfcalc(model, x, out y);
340        yield return y[0];
341      }
342    }
343  }
344}
Note: See TracBrowser for help on using the repository browser.