public static Network CreateNetworkMaxout(GPUModule module, int minibatchSize)
{
var net = new Network(module, minibatchSize: minibatchSize);
net.AddInputLayer(Constants.TOTAL_VALUE_COUNT, sparseDataSize: minibatchSize * RawRecord.FEATURE_COUNT * 2);
net.AddLabelLayer(1);
var fc1 = net.AddFullyConnectedLayer(128, "FC1");
fc1.Weights.InitValuesUniformCPU(0.1f);
fc1.L2Regularization = 0.00001f;
fc1.RegularizationRatio = 10;
net.AddMaxoutLayer("MAXOUT1", groupsize: 4);
var fc2 = net.AddFullyConnectedLayer(256, "FC2");
fc2.Weights.InitValuesUniformCPU(0.1f);
net.AddMaxoutLayer("MAXOUT2", groupsize: 2);
net.AddDropoutLayer();
var sm = net.AddSoftmaxLayer(2, "SMAX");
sm.Weights.InitValuesUniformCPU(0.1f);
return(net);
}
static void Main(string[] args)
{
var config = new NetworkConfiguration(
CostFunctionType.Quadratic,
new FlatDecayer(0.5),
28, 1);
var network = new Network(config);
network.AddConvolutionalLayer(5, 5, LearningRateAnnealerType.Adagrad);
network.AddPoolingLayer(2);
network.AddDetectorLayer(ActivatorType.LogisticActivator);
network.AddConvolutionalLayer(5, 3, LearningRateAnnealerType.Adagrad);
network.AddPoolingLayer(2);
network.AddDetectorLayer(ActivatorType.LogisticActivator);
network.AddFullyConnectedLayer(5, ActivatorType.LogisticActivator, LearningRateAnnealerType.Adagrad);
network.AddFullyConnectedLayer(4, ActivatorType.LogisticActivator, LearningRateAnnealerType.Adagrad);
foreach (var item in MnistReader.ReadTrainingData())
{
var data = item.Data.ConvertPixels();
var input = new double[1, data.GetLength(0), data.GetLength(1)];
for (int i = 0; i < data.GetLength(0); i++)
{
for (int j = 0; j < data.GetLength(1); j++)
{
input[0, i, j] = data[i, j];
}
}
var error = network.TrainModel(input, GetTargetOutput(item.Label));
}
}
