public void Test_Digits_MultiClass_Classification_Autoencoded()
{
var digits = Data.Digit.GetTrainingDigits();
var descriptor = Model.Descriptor.Create <Data.Digit>();
var autogen = new Supervised.NeuralNetwork.Encoders.AutoencoderGenerator()
{
Descriptor = descriptor,
OutputFunction = new Math.Functions.Logistic(),
MaxIterations = 200
};
var encoder = autogen.Generate(digits) as Supervised.NeuralNetwork.Encoders.AutoencoderModel;
Func <IEnumerable <Data.Digit>, IEnumerable <Data.Digit> > encoding = (s) =>
s.Select(d =>
new Data.Digit()
{
Label = d.Label,
Reading = encoder.PredictSequence(d.Reading).ToArray()
});
var model = MultiClassLearner.Learn(new Supervised.Regression.LogisticRegressionGenerator()
{
Descriptor = descriptor,
Lambda = 0.1,
PolynomialFeatures = 0
},
encoding(digits), 0.8, 0.5);
double count = 0;
var tests = Data.Digit.GetTestDigits();
foreach (var test in encoding(tests))
{
var prediction = (int)model.Predict(test);
count += (test.Label == prediction ? 1.0 : 0.0);
}
double accuracy = count / (double)tests.Count();
Console.WriteLine($"Digit Test:\nOverall Accuracy => { accuracy }");
Assert.True(model.Accuracy >= 0.65);
}
public void Test_Digits_MultiClass_Classification_Autoencoded()
{
var digits = Data.Digit.GetTrainingDigits();
var descriptor = Model.Descriptor.Create<Data.Digit>();
var autogen = new Supervised.NeuralNetwork.Encoders.AutoencoderGenerator() {
Descriptor = descriptor,
OutputFunction = new Math.Functions.Logistic(),
MaxIterations = 200
};
var encoder = autogen.Generate(digits) as Supervised.NeuralNetwork.Encoders.AutoencoderModel;
Func<IEnumerable<Data.Digit>, IEnumerable<Data.Digit>> encoding = (s) =>
s.Select(d =>
new Data.Digit()
{
Label = d.Label,
Reading = encoder.PredictSequence(d.Reading).ToArray()
});
var model = MultiClassLearner.Learn(new Supervised.Regression.LogisticRegressionGenerator()
{
Descriptor = descriptor,
Lambda = 0.1,
PolynomialFeatures = 0
},
encoding(digits), 0.8, 0.5);
double count = 0;
var tests = Data.Digit.GetTestDigits();
foreach (var test in encoding(tests))
{
var prediction = (int)model.Predict(test);
count += (test.Label == prediction ? 1.0 : 0.0);
}
double accuracy = count / (double)tests.Count();
Console.WriteLine($"Digit Test:\nOverall Accuracy => { accuracy }");
Assert.True(model.Accuracy >= 0.65);
}