Changeset 14310 for branches/HeuristicLab.Problems.GeneticProgramming.BloodGlucosePrediction/Problem.cs

Timestamp:

09/27/16 16:30:54 (8 years ago)

Author:

gkronber

Message:

changed interpreter and grammar to smooth ch and ins uptake using a compartement model

File:

• branches/HeuristicLab.Problems.GeneticProgramming.BloodGlucosePrediction/Problem.cs (modified) (6 diffs)

branches/HeuristicLab.Problems.GeneticProgramming.BloodGlucosePrediction/Problem.cs

@@ -21 +21 @@
 
 using System;
+using System.Collections;
 using System.Collections.Generic;
 using System.Linq;
+using HeuristicLab.Analysis;
 using HeuristicLab.Common;
 using HeuristicLab.Core;
@@ -62 +64 @@
 
     public override bool Maximization {
-      get { return false; }
+      get { return true; }
     }
 
@@ -98 +100 @@
       var rows = problemData.TrainingIndices.ToArray();
       var target = problemData.Dataset.GetDoubleValues(problemData.TargetVariable, rows);
-      var predicted = Interpreter.Apply(tree.Root.GetSubtree(0).GetSubtree(0), problemData.Dataset, rows);  
-
+      var predicted0 = Interpreter.Apply(tree.Root.GetSubtree(0).GetSubtree(0).GetSubtree(0), problemData.Dataset, rows);
+      var predicted1 = Interpreter.Apply(tree.Root.GetSubtree(0).GetSubtree(0).GetSubtree(1), problemData.Dataset, rows);
+      var predicted2 = Interpreter.Apply(tree.Root.GetSubtree(0).GetSubtree(0).GetSubtree(2), problemData.Dataset, rows);
+
+      var pred0_rsq = Rsq(predicted0, target);
+      var pred1_rsq = Rsq(predicted1, target);
+      var pred2_rsq = Rsq(predicted2, target);
+      return pred0_rsq + pred1_rsq + pred2_rsq;
+    }
+
+    private double Rsq(IEnumerable<double> predicted, IEnumerable<double> target) {
       // only take predictions for which the target is not NaN
       var selectedTuples = target.Zip(predicted, Tuple.Create).Where(t => !double.IsNaN(t.Item1)).ToArray();
@@ -106 +117 @@
 
       OnlineCalculatorError errorState;
-      var mse = OnlineMeanSquaredErrorCalculator.Calculate(target, predicted, out errorState);
-      if (errorState != OnlineCalculatorError.None) mse = 1E6;
-      return mse;
-    }
-
-    public override void Analyze(ISymbolicExpressionTree[] trees, double[] qualities, ResultCollection results, IRandom random) {
+      var r = OnlinePearsonsRCalculator.Calculate(target, predicted, out errorState);
+      if (errorState != OnlineCalculatorError.None) r = 0;
+      return r * r;
+    }
+
+    public override void Analyze(ISymbolicExpressionTree[] trees, double[] qualities, ResultCollection results,
+      IRandom random) {
       base.Analyze(trees, qualities, results, random);
 
       if (!results.ContainsKey("Solution")) {
         results.Add(new Result("Solution", typeof(IRegressionSolution)));
+      }
+      if (!results.ContainsKey("Terms")) {
+        results.Add(new Result("Terms", typeof(DataTable)));
       }
 
@@ -121 +136 @@
       var bestQuality = qualities.First();
       for (int i = 1; i < trees.Length; i++) {
-        if (qualities[i] < bestQuality) {
+        if (qualities[i] > bestQuality) {
           bestQuality = qualities[i];
           bestTree = trees[i];
@@ -127 +142 @@
       }
 
+
       var clonedProblemData = (IRegressionProblemData)ProblemData.Clone();
-      var model = new Model(clonedProblemData, (ISymbolicExpressionTree)bestTree.Clone());
-      results["Solution"].Value = model.CreateRegressionSolution(clonedProblemData);
+      var rows = clonedProblemData.TrainingIndices.ToArray();
+      var target = clonedProblemData.Dataset.GetDoubleValues(clonedProblemData.TargetVariable, rows).ToArray();
+      var predicted0 =
+        Interpreter.Apply(bestTree.Root.GetSubtree(0).GetSubtree(0).GetSubtree(0), clonedProblemData.Dataset, rows)
+          .ToArray();
+      var predicted1 =
+        Interpreter.Apply(bestTree.Root.GetSubtree(0).GetSubtree(0).GetSubtree(1), clonedProblemData.Dataset, rows)
+          .ToArray();
+      var predicted2 =
+        Interpreter.Apply(bestTree.Root.GetSubtree(0).GetSubtree(0).GetSubtree(2), clonedProblemData.Dataset, rows)
+          .ToArray();
+
+      var termsTable = new HeuristicLab.Analysis.DataTable("Terms");
+      var r0 = new DataRow("GlucTerm", "GlucTerm", predicted0);
+      var r1 = new DataRow("InsTerm", "InsTerm", predicted1);
+      r1.VisualProperties.SecondYAxis = true;
+      var r2 = new DataRow("ChTerm", "ChTerm", predicted2);
+      r2.VisualProperties.SecondYAxis = true;
+      var r3 = new DataRow("Target", "Target", target);
+      termsTable.Rows.Add(r0);
+      termsTable.Rows.Add(r1);
+      termsTable.Rows.Add(r2);
+      termsTable.Rows.Add(r3);
+      results["Terms"].Value = termsTable;
+
+
+      var filteredPredicted0 = rows.Where(r => !double.IsNaN(target[r])).Select(r => predicted0[r]).ToArray();
+      var filteredPredicted1 = rows.Where(r => !double.IsNaN(target[r])).Select(r => predicted1[r]).ToArray();
+      var filteredPredicted2 = rows.Where(r => !double.IsNaN(target[r])).Select(r => predicted2[r]).ToArray();
+      var filteredTarget = target.Where(t => !double.IsNaN(t)).ToArray();
+
+
+      var ds = new ModifiableDataset(new string[] { "pred0", "pred1", "pred2", "target" },
+        new List<IList>
+        {
+          filteredPredicted0.ToList(),
+          filteredPredicted1.ToList(),
+          filteredPredicted2.ToList(),
+          filteredTarget.ToList()
+        });
+      var lrProbData = new RegressionProblemData(ds, new string[] { "pred0", "pred1", "pred2" }, "target");
+      lrProbData.TrainingPartition.Start = clonedProblemData.TrainingPartition.Start;
+      lrProbData.TrainingPartition.End = clonedProblemData.TrainingPartition.End;
+      lrProbData.TestPartition.Start = clonedProblemData.TestPartition.Start;
+      lrProbData.TestPartition.End = clonedProblemData.TestPartition.End;
+
+      try {
+        double rmsError, cvRmsError;
+        var lrSolution = HeuristicLab.Algorithms.DataAnalysis.LinearRegression.CreateLinearRegressionSolution(
+          lrProbData, out rmsError, out cvRmsError);
+        results["Solution"].Value = lrSolution;
+      } catch (Exception) {
+        // ignore
+      }
     }