Changeset 8473 for trunk/sources/HeuristicLab.Tests/HeuristicLab.Algorithms.DataAnalysis-3.4/GaussianProcessModelTest.cs

Timestamp:

08/10/12 20:42:30 (12 years ago)

Author:

gkronber

Message:

#1902 worked on GPR: added line chart, made parameters of mean and covariance functions readable, removed target variable scaling, moved noise hyperparameter for likelihood function to the end of the parameter list, added methods to calculate the predicted variance, removed limits for scale of covariance functions and introduced exception handling to catch non-spd or singular cov matrixes, implemented rational quadratic covariance function, added unit test case from GBML book (however it does not work as the book seemingly uses a noise-less likelihood function)

File:

• trunk/sources/HeuristicLab.Tests/HeuristicLab.Algorithms.DataAnalysis-3.4/GaussianProcessModelTest.cs (modified) (3 diffs)

trunk/sources/HeuristicLab.Tests/HeuristicLab.Algorithms.DataAnalysis-3.4/GaussianProcessModelTest.cs

@@ -56 +56 @@
 
         var dHyp = model.GetHyperparameterGradients();
-        Assert.AreEqual(-2.0171e+003, dHyp[0], 1);
-        Assert.AreEqual(-248.7932, dHyp[1], 1E-2);
+        Assert.AreEqual(-248.7932, dHyp[0], 1E-2);
         var dHypCovExpected = new double[] { -0.5550e4, -5.5533e4, -0.2511e4, -2.7625e4, -1.3033e4, 0.0289e4, -2.7625e4 };
-        AssertEqual(dHypCovExpected, dHyp.Skip(2).ToArray(), 1);
+        AssertEqual(dHypCovExpected, dHyp.Skip(1).Take(7).ToArray(), 1);
+        Assert.AreEqual(-2.0171e+003, dHyp.Last(), 1);
+
 
         var predTrain = model.GetEstimatedValues(problemData.Dataset, new int[] { 0, 400 }).ToArray();
         Assert.AreEqual(310.5930, predTrain[0], 1e-3);
         Assert.AreEqual(347.9993, predTrain[1], 1e-3);
+
+        var predTrainVar = model.GetEstimatedVariance(problemData.Dataset, problemData.TrainingIndices).ToArray();
       }
 
       {
-        var hyp = new double[] { 0.716427415979145, 0.029973094285941, 0.455535210579926, 3.438647883940457, 1.464114485889487, 3.001788584487478, 3.815289323309630, 4.374914122810222, 3.001788584487478 };
+        var hyp = new double[] { 0.029973094285941, 0.455535210579926, 3.438647883940457, 1.464114485889487, 3.001788584487478, 3.815289323309630, 4.374914122810222, 3.001788584487478, 0.716427415979145 };
         var model = new GaussianProcessModel(problemData.Dataset, targetVariable, allowedInputVariables, rows, hyp,
                                              meanFunction,
@@ -74 +77 @@
 
         var dHyp = model.GetHyperparameterGradients();
-        Assert.AreEqual(0.8621, dHyp[0], 1e-3);
-        Assert.AreEqual(-0.0046, dHyp[1], 1e-3);
+        Assert.AreEqual(-0.0046, dHyp[0], 1e-3);
         var dHypCovExpected = new double[] { 0.2652, -0.2386, 0.1706, -0.1744, 0.0000, 0.0000, -0.1744 };
-        AssertEqual(dHypCovExpected, dHyp.Skip(2).ToArray(), 1e-3);
+        AssertEqual(dHypCovExpected, dHyp.Skip(1).Take(7).ToArray(), 1e-3);
+        Assert.AreEqual(0.8621, dHyp.Last(), 1e-3);
 
         var predTrain = model.GetEstimatedValues(problemData.Dataset, new int[] { 0, 400 }).ToArray();
@@ -83 +86 @@
         Assert.AreEqual(356.6076, predTrain[1], 1e-3);
       }
+
+      /*
+      {
+        // example from GPML book
+        var hyp = new double[] { 
+          341.0, // mean 341 ppm
+          // SE iso (long term trend)
+          Math.Log(67.0 / 45.0), // length scale 67 years
+          Math.Log(Math.Sqrt(66)), // magnitude 66ppm
+                    
+          // product of SEiso and periodic
+          Math.Log(90.0 / 45.0), // decay-time 90 years
+          Math.Log(Math.Sqrt(2.4)), // magnitude 2.4ppm
+
+          Math.Log(1.3), // smoothness
+          Math.Log(1), // period 1 year
+          Math.Log(Math.Sqrt(2.4)), // magnitude 2.4ppm
+
+          // short term variation
+          Math.Log(1.2 / 45.0), // typical length 1.2 years
+          Math.Log(Math.Sqrt(0.66)), // magnitude 0.66ppm
+          Math.Log(0.78), // shape (very small)
+
+          // SEiso (correlated noise)
+          Math.Log(1.6 / 45.0 / 12.0), // 1.6 months
+          Math.Log(Math.Sqrt(0.18)), // amplitude of correlated noise 0.18ppm
+          Math.Log(Math.Sqrt(0.19)),  // theta11 0.19ppm noise
+          };
+
+        covarianceFunction.Terms.Add(new CovarianceRQiso());
+        covarianceFunction.Terms.Add(new CovarianceSEiso()); // correlated noise
+        var model = new GaussianProcessModel(problemData.Dataset, targetVariable, allowedInputVariables, Enumerable.Range(0, 545), hyp,
+                                             new MeanConst(),
+                                             covarianceFunction);
+        Assert.AreEqual(-108.5, model.NegativeLogLikelihood, 1);
+      }
+       */
     }