static void MNISTLeakyReLUSigmoidAdamAndMeanSquaredError()
{
ConsoleUI.WriteLine("LeakyReLU, Sigmoid - MSE, Adam");
var(XTrain, yTrain) = MNISTHelper.LoadTraining(0, 1);
var(XTest, yTest) = MNISTHelper.LoadTesting(0, 1);
var model = new Model(new AdamOptimiser(learningRate: 2e-3, beta1: 0.5), new MeanSquaredErrorCost());
model.Add(new DenseLayer(28 * 28, 100, new LeakyReLUActivation(0.05)));
model.Add(new DenseLayer(100, 10, new SigmoidActivation()));
MNISTTest(model, XTrain, yTrain, XTest, yTest, 1);
}
static void MNISTLTanHSigmoidSDGAndMeanSquareError()
{
ConsoleUI.WriteLine("TanH, Sigmoid - MSE, SGD");
var(XTrain, yTrain) = MNISTHelper.LoadTraining(-1, 1);
var(XTest, yTest) = MNISTHelper.LoadTesting(-1, 1);
var model = new Model(new SGDOptimiser(learningRate: 0.1), new MeanSquaredErrorCost());
model.Add(new DenseLayer(28 * 28, 100, new TanhActivation()));
model.Add(new DenseLayer(100, 10, new SigmoidActivation()));
MNISTTest(model, XTrain, yTrain, XTest, yTest, 1);
}
private static void SimpleGAN()
{
string imageFolder = CreateImageFolder();
ConsoleUI.WriteLine("Enter a comma separated list of numbers from 0 to 9 to train the GAN on those digits.");
ConsoleUI.WriteLine("Just press enter to train the GAN on all digits (this will take a *long* time).");
ConsoleUI.WriteLine($"Example generated images will be written into '{imageFolder}' at the end of each epoch.");
var(XTrain, _) = MNISTHelper.LoadTraining(scaleMin: -1, scaleMax: 1, filter: GetFilterFromUser());
var adam = new AdamOptimiser(learningRate: 2e-4, beta1: 0.5, beta2: 0.999);
var generator = BuildGenerator(adam);
var discriminator = BuildDiscriminator(adam);
Train(imageFolder, XTrain, generator, discriminator);
}
