public void RNNPredictsWell()
{
QuqeUtil.Random = new Random(42);
var data = NNTestUtils.GetData("2004-01-01", "2004-07-01");
var trainResult = Train(data, 8, 4, 1000);
var trainingFitness = Functions.ComputeFitness(new PredictorWithInputs(new RNN(trainResult.RNNSpec), data.Input), data);
Trace.WriteLine("Training fitness: " + trainingFitness);
trainingFitness.ShouldEqual(0.968);
}
public void RBFTrainingIsReproducible()
{
var data = NNTestUtils.GetData("2004-01-01", "2004-05-01");
var fitnesses = Lists.Repeat(5, i => {
var rbfNet = RBFNet.Train(data.Input, data.Output, 0.1, 1);
rbfNet.IsDegenerate.ShouldBeFalse();
return(Functions.ComputeFitness(new PredictorWithInputs(rbfNet, data.Input), data));
});
fitnesses.Distinct().Count().ShouldEqual(1);
}
public void RNNTrainsQuickly()
{
var sw = new Stopwatch();
sw.Start();
var trainingData = NNTestUtils.GetData("2004-01-01", "2005-01-01");
Train(trainingData, 32, 16, 1000);
sw.Stop();
sw.ElapsedMilliseconds.ShouldBeGreaterThan(2000).ShouldBeLessThan(17000);
}
public void RNNTrainingIsReproducible()
{
var data = NNTestUtils.GetData("2004-01-01", "2004-05-01");
var checksums = Lists.Repeat(5, i => {
QuqeUtil.Random = new Random(42);
var trainResult = Train(data, 8, 4, 1000);
return(NNTestUtils.Checksum(trainResult.RNNSpec.Weights));
});
checksums.Distinct().Count().ShouldEqual(1);
}
public void RBFPredictsWell()
{
var data = NNTestUtils.GetData("2004-01-01", "2004-05-01");
Func <double, double> fitnessForTolerance = tolerance => {
var rbfNet = RBFNet.Train(data.Input, data.Output, tolerance, 1);
rbfNet.IsDegenerate.ShouldBeFalse();
var fitness = Functions.ComputeFitness(new PredictorWithInputs(rbfNet, data.Input), data);
Trace.WriteLine("RBF fitness: " + fitness);
return(fitness);
};
fitnessForTolerance(0.2).ShouldBeGreaterThan(0.9);
fitnessForTolerance(0.01).ShouldEqual(1.0);
}
