public void test_constructor_fitness_R2Adjusted()
{
/*Test that constructor compiles with interface implementation as parameter, and
* fitness parameter is of type AdjustedR2.*/
Fitness adjR2 = new AdjustedR2();
Variable y = new Variable("", new List <double>());
Model m = new Model(y, new List <Variable>());
FullSelection sel = new FullSelection(m, adjR2);
Assert.True(sel.fitness.GetType() == typeof(AdjustedR2));
}
public void test_SelectBestFit_performance_10000_observations_6_variables()
{
/*Test is done with a file containing 100 000 observations with 6 variables. */
Directory.SetCurrentDirectory(Path.Combine(TestContext.CurrentContext.TestDirectory, @"..\..\"));
Model m = CSV.ToModel(@"test_files\performance_test.csv", "ord.num");
Fitness fitness = new AdjustedR2();
FullSelection sel = new FullSelection(m, fitness);
Stopwatch sw = new Stopwatch();
sw.Start();
sel.SelectBestFit();
sw.Stop();
Console.WriteLine("took + " + sw.ElapsedMilliseconds + " milliseconds");
Assert.LessOrEqual(sw.ElapsedMilliseconds, 10000);
}
private static void DoModelSelection()
{
Model model, results;
if (filepath == null)
{
Write.Error("No file read in");
return;
}
if (responseVariable == null)
{
Write.Error("No response variable selected");
return;
}
model = CSV.ToModel(filepath, responseVariable);
FullSelection sel = new FullSelection(model, fitness);
results = sel.SelectBestFit();
Write.Model(results);
}
public void test_SelectBestFit_throws_MathError()
{
/*Matrix needs to be positive definite for using MathNets Fit.MultiDim()-function.
* This should fail and throw ArgumentException. This is noticed after the threads
* have finished because bestModels is be 0-length. Then MathError is thrown. */
List <double> d1 = new List <double>();
List <double> d2 = new List <double>();
List <double> d3 = new List <double>();
List <double> d4 = new List <double>();
List <double> d5 = new List <double>();
List <double> d6 = new List <double>();
for (int i = 0; i < 1000; i++)
{
d1.Add(1);
d2.Add(1);
d3.Add(1);
d4.Add(1);
d5.Add(1);
d6.Add(1);
}
Variable yList = new Variable("list", d1);
Variable list2 = new Variable("list2", d2);
Variable list3 = new Variable("list3", d3);
Variable list4 = new Variable("list4", d4);
Variable list5 = new Variable("list5", d5);
Variable list6 = new Variable("list6", d6);
List <Variable> x = new List <Variable>()
{
list2, list3, list4, list5, list6
};
Fitness adjR2 = new AdjustedR2();
Model m = new Model(yList, x);
FullSelection sel = new FullSelection(m, adjR2);
Assert.Throws <MathError>(() => sel.SelectBestFit());
}
public void test_SelectBestFit()
{
/*R used to verify results: best fitted model based on adjusted coefficient of
* determination contains variables list3 and list4, with adjusted R^2 value of 0.6721534.*/
Variable yList = new Variable("list", new List <double>()
{
2.2, 3.5, 3, 14, 8, 2
});
Variable list2 = new Variable("list2", new List <double>()
{
3, 15.2, 1.1, 2, 3, 2
});
Variable list3 = new Variable("list3", new List <double>()
{
1, 2, 3, 4, 5, 2.2
});
Variable list4 = new Variable("list4", new List <double>()
{
1, 1.1, 1.4, 1.3, 1.5, 1.2
});
List <Variable> xList = new List <Variable>()
{
list2, list3, list4
};
Fitness adjR2 = new AdjustedR2();
Model m = new Model(yList, xList);
FullSelection sel = new FullSelection(m, adjR2);
Model bestFit = sel.SelectBestFit();
Assert.AreEqual(0.6721534, Math.Round(bestFit.fitness, 7));
Assert.Contains(list3, bestFit.getXVars());
Assert.Contains(list4, bestFit.getXVars());
Assert.False(bestFit.getXVars().Contains(list2));
}
