static void Main(string[] args)
{
Torch.SetSeed(1);
var device = Torch.IsCudaAvailable() ? Device.CUDA : Device.CPU;
if (device.Type == DeviceType.CUDA)
{
_trainBatchSize *= 8;
_testBatchSize *= 8;
_epochs *= 16;
}
Console.WriteLine();
Console.WriteLine($"\tRunning AlexNet with {_dataset} on {device.Type.ToString()} for {_epochs} epochs");
Console.WriteLine();
var sourceDir = _dataLocation;
var targetDir = Path.Combine(_dataLocation, "test_data");
if (!Directory.Exists(targetDir))
{
Directory.CreateDirectory(targetDir);
Utils.Decompress.ExtractTGZ(Path.Combine(sourceDir, "cifar-10-binary.tar.gz"), targetDir);
}
using (var train = new CIFARReader(targetDir, false, _trainBatchSize, shuffle: true, device: device))
using (var test = new CIFARReader(targetDir, true, _testBatchSize, device: device))
using (var model = new Model("model", _numClasses, device))
using (var optimizer = NN.Optimizer.Adam(model.parameters(), 0.001)) {
Stopwatch sw = new Stopwatch();
sw.Start();
for (var epoch = 1; epoch <= _epochs; epoch++)
{
Train(model, optimizer, nll_loss(), train, epoch, _trainBatchSize, train.Size);
Test(model, nll_loss(), test, test.Size);
GC.Collect();
if (sw.Elapsed.TotalSeconds > 3600)
{
break;
}
}
sw.Stop();
Console.WriteLine($"Elapsed time {sw.Elapsed.TotalSeconds} s.");
Console.ReadLine();
}
}
private readonly static int _timeout = 3600; // One hour by default.
internal static void Main(string[] args)
{
torch.random.manual_seed(1);
var device =
// This worked on a GeForce RTX 2080 SUPER with 8GB, for all the available network architectures.
// It may not fit with less memory than that, but it's worth modifying the batch size to fit in memory.
torch.cuda.is_available() ? torch.CUDA :
torch.CPU;
if (device.type == DeviceType.CUDA)
{
_trainBatchSize *= 8;
_testBatchSize *= 8;
_epochs *= 16;
}
var modelName = args.Length > 0 ? args[0] : "AlexNet";
var epochs = args.Length > 1 ? int.Parse(args[1]) : _epochs;
var timeout = args.Length > 2 ? int.Parse(args[2]) : _timeout;
Console.WriteLine();
Console.WriteLine($"\tRunning {modelName} with {_dataset} on {device.type.ToString()} for {epochs} epochs, terminating after {TimeSpan.FromSeconds(timeout)}.");
Console.WriteLine();
var sourceDir = _dataLocation;
var targetDir = Path.Combine(_dataLocation, "test_data");
if (!Directory.Exists(targetDir))
{
Directory.CreateDirectory(targetDir);
Utils.Decompress.ExtractTGZ(Path.Combine(sourceDir, "cifar-10-binary.tar.gz"), targetDir);
}
Console.WriteLine($"\tCreating the model...");
Module model = null;
switch (modelName.ToLower())
{
case "alexnet":
model = new AlexNet(modelName, _numClasses, device);
break;
case "mobilenet":
model = new MobileNet(modelName, _numClasses, device);
break;
case "vgg11":
case "vgg13":
case "vgg16":
case "vgg19":
model = new VGG(modelName, _numClasses, device);
break;
case "resnet18":
model = ResNet.ResNet18(_numClasses, device);
break;
case "resnet34":
_testBatchSize /= 4;
model = ResNet.ResNet34(_numClasses, device);
break;
case "resnet50":
_trainBatchSize /= 6;
_testBatchSize /= 8;
model = ResNet.ResNet50(_numClasses, device);
break;
#if false
// The following is disabled, because they require big CUDA processors in order to run.
case "resnet101":
_trainBatchSize /= 6;
_testBatchSize /= 8;
model = ResNet.ResNet101(_numClasses, device);
break;
case "resnet152":
_testBatchSize /= 4;
model = ResNet.ResNet152(_numClasses, device);
break;
#endif
}
var hflip = torchvision.transforms.HorizontalFlip();
var gray = torchvision.transforms.Grayscale(3);
var rotate = torchvision.transforms.Rotate(90);
var contrast = torchvision.transforms.AdjustContrast(1.25);
Console.WriteLine($"\tPreparing training and test data...");
Console.WriteLine();
using (var train = new CIFARReader(targetDir, false, _trainBatchSize, shuffle: true, device: device, transforms: new torchvision.ITransform[] { }))
using (var test = new CIFARReader(targetDir, true, _testBatchSize, device: device))
using (var optimizer = torch.optim.Adam(model.parameters(), 0.001)) {
Stopwatch totalSW = new Stopwatch();
totalSW.Start();
for (var epoch = 1; epoch <= epochs; epoch++)
{
Stopwatch epchSW = new Stopwatch();
epchSW.Start();
Train(model, optimizer, nll_loss(), train.Data(), epoch, _trainBatchSize, train.Size);
Test(model, nll_loss(), test.Data(), test.Size);
epchSW.Stop();
Console.WriteLine($"Elapsed time for this epoch: {epchSW.Elapsed.TotalSeconds} s.");
if (totalSW.Elapsed.TotalSeconds > timeout)
{
break;
}
}
totalSW.Stop();
Console.WriteLine($"Elapsed training time: {totalSW.Elapsed} s.");
}
model.Dispose();
}