private static void Train(string baseName, string dataset, uint epoch, double learningRate, double minLearningRate, uint miniBatchSize, uint validation, bool useMean)
{
try
{
IList <Matrix <RgbPixel> > trainingImages;
IList <uint> trainingLabels;
IList <Matrix <RgbPixel> > testingImages;
IList <uint> testingLabels;
var mean = useMean ? Path.Combine(dataset, "train.mean.bmp") : null;
Console.WriteLine("Start load train images");
Load("train", dataset, mean, out trainingImages, out trainingLabels);
Console.WriteLine($"Load train images: {trainingImages.Count}");
Console.WriteLine("Start load test images");
Load("test", dataset, mean, out testingImages, out testingLabels);
Console.WriteLine($"Load test images: {testingImages.Count}");
// So with that out of the way, we can make a network instance.
var trainNet = NativeMethods.LossMulticlassLog_age_train_type_create();
var networkId = LossMulticlassLogRegistry.GetId(trainNet);
LossMulticlassLogRegistry.Add(trainNet);
using (var net = new LossMulticlassLog(networkId))
using (var trainer = new DnnTrainer <LossMulticlassLog>(net))
{
trainer.SetLearningRate(learningRate);
trainer.SetMinLearningRate(minLearningRate);
trainer.SetMiniBatchSize(miniBatchSize);
trainer.BeVerbose();
trainer.SetSynchronizationFile(baseName, 180);
// create array box
var trainingImagesCount = trainingImages.Count;
var trainingLabelsCount = trainingLabels.Count;
var maxIteration = (int)Math.Ceiling(trainingImagesCount / (float)miniBatchSize);
var imageBatches = new Matrix <RgbPixel> [maxIteration][];
var labelBatches = new uint[maxIteration][];
for (var i = 0; i < maxIteration; i++)
{
if (miniBatchSize <= trainingImagesCount - i * miniBatchSize)
{
imageBatches[i] = new Matrix <RgbPixel> [miniBatchSize];
labelBatches[i] = new uint[miniBatchSize];
}
else
{
imageBatches[i] = new Matrix <RgbPixel> [trainingImagesCount % miniBatchSize];
labelBatches[i] = new uint[trainingLabelsCount % miniBatchSize];
}
}
using (var fs = new FileStream($"{baseName}.log", FileMode.Create, FileAccess.Write, FileShare.Write))
using (var sw = new StreamWriter(fs, Encoding.UTF8))
for (var e = 0; e < epoch; e++)
{
var randomArray = Enumerable.Range(0, trainingImagesCount).OrderBy(i => Guid.NewGuid()).ToArray();
var index = 0;
for (var i = 0; i < imageBatches.Length; i++)
{
var currentImages = imageBatches[i];
var currentLabels = labelBatches[i];
for (var j = 0; j < imageBatches[i].Length; j++)
{
var rIndex = randomArray[index];
currentImages[j] = trainingImages[rIndex];
currentLabels[j] = trainingLabels[rIndex];
index++;
}
}
for (var i = 0; i < maxIteration; i++)
{
LossMulticlassLog.TrainOneStep(trainer, imageBatches[i], labelBatches[i]);
}
var lr = trainer.GetLearningRate();
var loss = trainer.GetAverageLoss();
var trainLog = $"Epoch: {e}, learning Rate: {lr}, average loss: {loss}";
Console.WriteLine(trainLog);
sw.WriteLine(trainLog);
if (e > 0 && e % validation == 0)
{
Validation(baseName, net, trainingImages, trainingLabels, testingImages, testingLabels, false, false, out var trainAccuracy, out var testAccuracy);
var validationLog = $"Epoch: {e}, train accuracy: {trainAccuracy}, test accuracy: {testAccuracy}";
Console.WriteLine(validationLog);
sw.WriteLine(validationLog);
}
if (lr < minLearningRate)
{
break;
}
}
// wait for training threads to stop
trainer.GetNet();
Console.WriteLine("done training");
net.Clean();
LossMulticlassLog.Serialize(net, $"{baseName}.dat");
// Now let's run the training images through the network. This statement runs all the
// images through it and asks the loss layer to convert the network's raw output into
// labels. In our case, these labels are the numbers between 0 and 9.
Validation(baseName, net, trainingImages, trainingLabels, testingImages, testingLabels, true, true, out _, out _);
}
}
catch (Exception e)
{
Console.WriteLine(e);
}
}
private static int Main(string[] args)
{
if (args.Length != 1)
{
Console.WriteLine("This example needs the MNIST dataset to run!");
Console.WriteLine("You can get MNIST from http://yann.lecun.com/exdb/mnist/");
Console.WriteLine("Download the 4 files that comprise the dataset, decompress them, and");
Console.WriteLine("put them in a folder. Then give that folder as input to this program.");
return(1);
}
try
{
// MNIST is broken into two parts, a training set of 60000 images and a test set of
// 10000 images. Each image is labeled so that we know what hand written digit is
// depicted. These next statements load the dataset into memory.
IList <Matrix <byte> > trainingImages;
IList <uint> trainingLabels;
IList <Matrix <byte> > testingImages;
IList <uint> testingLabels;
Dlib.LoadMNISTDataset(args[0], out trainingImages, out trainingLabels, out testingImages, out testingLabels);
// Now let's define the LeNet. Broadly speaking, there are 3 parts to a network
// definition. The loss layer, a bunch of computational layers, and then an input
// layer. You can see these components in the network definition below.
//
// The input layer here says the network expects to be given matrix<unsigned char>
// objects as input. In general, you can use any dlib image or matrix type here, or
// even define your own types by creating custom input layers.
//
// Then the middle layers define the computation the network will do to transform the
// input into whatever we want. Here we run the image through multiple convolutions,
// ReLU units, max pooling operations, and then finally a fully connected layer that
// converts the whole thing into just 10 numbers.
//
// Finally, the loss layer defines the relationship between the network outputs, our 10
// numbers, and the labels in our dataset. Since we selected loss_multiclass_log it
// means we want to do multiclass classification with our network. Moreover, the
// number of network outputs (i.e. 10) is the number of possible labels. Whichever
// network output is largest is the predicted label. So for example, if the first
// network output is largest then the predicted digit is 0, if the last network output
// is largest then the predicted digit is 9.
// This net_type defines the entire network architecture. For example, the block
// relu<fc<84,SUBNET>> means we take the output from the subnetwork, pass it through a
// fully connected layer with 84 outputs, then apply ReLU. Similarly, a block of
// max_pool<2,2,2,2,relu<con<16,5,5,1,1,SUBNET>>> means we apply 16 convolutions with a
// 5x5 filter size and 1x1 stride to the output of a subnetwork, then apply ReLU, then
// perform max pooling with a 2x2 window and 2x2 stride.
// So with that out of the way, we can make a network instance.
using (var net = new LossMulticlassLog(3))
{
// And then train it using the MNIST data. The code below uses mini-batch stochastic
// gradient descent with an initial learning rate of 0.01 to accomplish this.
using (var trainer = new DnnTrainer <LossMulticlassLog>(net))
{
trainer.SetLearningRate(0.01);
trainer.SetMinLearningRate(0.00001);
trainer.SetMiniBatchSize(128);
trainer.BeVerbose();
// Since DNN training can take a long time, we can ask the trainer to save its state to
// a file named "mnist_sync" every 20 seconds. This way, if we kill this program and
// start it again it will begin where it left off rather than restarting the training
// from scratch. This is because, when the program restarts, this call to
// set_synchronization_file() will automatically reload the settings from mnist_sync if
// the file exists.
trainer.SetSynchronizationFile("mnist_sync", 20);
// Finally, this line begins training. By default, it runs SGD with our specified
// learning rate until the loss stops decreasing. Then it reduces the learning rate by
// a factor of 10 and continues running until the loss stops decreasing again. It will
// keep doing this until the learning rate has dropped below the min learning rate
// defined above or the maximum number of epochs as been executed (defaulted to 10000).
LossMulticlassLog.Train(trainer, trainingImages, trainingLabels);
// At this point our net object should have learned how to classify MNIST images. But
// before we try it out let's save it to disk. Note that, since the trainer has been
// running images through the network, net will have a bunch of state in it related to
// the last batch of images it processed (e.g. outputs from each layer). Since we
// don't care about saving that kind of stuff to disk we can tell the network to forget
// about that kind of transient data so that our file will be smaller. We do this by
// "cleaning" the network before saving it.
net.Clean();
LossMulticlassLog.Serialize(net, "mnist_network.dat");
// Now if we later wanted to recall the network from disk we can simply say:
// deserialize("mnist_network.dat") >> net;
// Now let's run the training images through the network. This statement runs all the
// images through it and asks the loss layer to convert the network's raw output into
// labels. In our case, these labels are the numbers between 0 and 9.
using (var predictedLabels = net.Operator(trainingImages))
{
var numRight = 0;
var numWrong = 0;
// And then let's see if it classified them correctly.
for (var i = 0; i < trainingImages.Count; ++i)
{
if (predictedLabels[i] == trainingLabels[i])
{
++numRight;
}
else
{
++numWrong;
}
}
Console.WriteLine($"training num_right: {numRight}");
Console.WriteLine($"training num_wrong: {numWrong}");
Console.WriteLine($"training accuracy: {numRight / (double)(numRight + numWrong)}");
// Let's also see if the network can correctly classify the testing images. Since
// MNIST is an easy dataset, we should see at least 99% accuracy.
using (var predictedLabels2 = net.Operator(testingImages))
{
numRight = 0;
numWrong = 0;
for (var i = 0; i < testingImages.Count; ++i)
{
if (predictedLabels2[i] == testingLabels[i])
{
++numRight;
}
else
{
++numWrong;
}
}
Console.WriteLine($"testing num_right: {numRight}");
Console.WriteLine($"testing num_wrong: {numWrong}");
Console.WriteLine($"testing accuracy: {numRight / (double)(numRight + numWrong)}");
// Finally, you can also save network parameters to XML files if you want to do
// something with the network in another tool. For example, you could use dlib's
// tools/convert_dlib_nets_to_caffe to convert the network to a caffe model.
Dlib.NetToXml(net, "lenet.xml");
}
}
}
}
}
catch (Exception e)
{
Console.WriteLine(e.Message);
}
return(0);
}
private void Train(Parameter parameter)
{
try
{
IList <Matrix <C> > trainingImages;
IList <T> trainingLabels;
IList <Matrix <C> > testingImages;
IList <T> testingLabels;
Logger.Info("Start load train images");
Load(parameter.Dataset, "train", out trainingImages, out trainingLabels);
Logger.Info($"Load train images: {trainingImages.Count}");
Logger.Info("Start load test images");
Load(parameter.Dataset, "test", out testingImages, out testingLabels);
Logger.Info($"Load test images: {testingImages.Count}");
Logger.Info("");
// So with that out of the way, we can make a network instance.
var networkId = SetupNetwork();
using (var net = new LossMulticlassLog(networkId))
using (var solver = new Adam())
using (var trainer = new DnnTrainer <LossMulticlassLog>(net, solver))
{
var learningRate = parameter.LearningRate;
var minLearningRate = parameter.MinLearningRate;
var miniBatchSize = parameter.MiniBatchSize;
var baseName = parameter.BaseName;
var epoch = parameter.Epoch;
var validation = parameter.Validation;
trainer.SetLearningRate(learningRate);
trainer.SetMinLearningRate(minLearningRate);
trainer.SetMiniBatchSize(miniBatchSize);
trainer.BeVerbose();
trainer.SetSynchronizationFile(baseName, 180);
// create array box
var trainingImagesCount = trainingImages.Count;
var trainingLabelsCount = trainingLabels.Count;
var maxIteration = (int)Math.Ceiling(trainingImagesCount / (float)miniBatchSize);
var imageBatches = new Matrix <C> [maxIteration][];
var labelBatches = new uint[maxIteration][];
for (var i = 0; i < maxIteration; i++)
{
if (miniBatchSize <= trainingImagesCount - i * miniBatchSize)
{
imageBatches[i] = new Matrix <C> [miniBatchSize];
labelBatches[i] = new uint[miniBatchSize];
}
else
{
imageBatches[i] = new Matrix <C> [trainingImagesCount % miniBatchSize];
labelBatches[i] = new uint[trainingLabelsCount % miniBatchSize];
}
}
using (var fs = new FileStream($"{baseName}.log", FileMode.Create, FileAccess.Write, FileShare.Write))
using (var sw = new StreamWriter(fs, Encoding.UTF8))
for (var e = 0; e < epoch; e++)
{
var randomArray = Enumerable.Range(0, trainingImagesCount).OrderBy(i => Guid.NewGuid()).ToArray();
var index = 0;
for (var i = 0; i < imageBatches.Length; i++)
{
var currentImages = imageBatches[i];
var currentLabels = labelBatches[i];
for (var j = 0; j < imageBatches[i].Length; j++)
{
var rIndex = randomArray[index];
currentImages[j] = trainingImages[rIndex];
currentLabels[j] = this.Cast(trainingLabels[rIndex]);
index++;
}
}
for (var i = 0; i < maxIteration; i++)
{
LossMulticlassLog.TrainOneStep(trainer, imageBatches[i], labelBatches[i]);
}
var lr = trainer.GetLearningRate();
var loss = trainer.GetAverageLoss();
var trainLog = $"Epoch: {e}, learning Rate: {lr}, average loss: {loss}";
Logger.Info(trainLog);
sw.WriteLine(trainLog);
if (e >= 0 && e % validation == 0)
{
var validationParameter = new ValidationParameter <T, C>
{
BaseName = parameter.BaseName,
Output = parameter.Output,
Trainer = net,
TrainingImages = trainingImages,
TrainingLabels = trainingLabels,
TestingImages = testingImages,
TestingLabels = testingLabels,
UseConsole = true,
SaveToXml = true,
OutputDiffLog = true
};
Validation(validationParameter, out var trainAccuracy, out var testAccuracy);
var validationLog = $"Epoch: {e}, train accuracy: {trainAccuracy}, test accuracy: {testAccuracy}";
Logger.Info(validationLog);
sw.WriteLine(validationLog);
var name = this.GetBaseName(parameter.Epoch,
parameter.LearningRate,
parameter.MinLearningRate,
parameter.MiniBatchSize);
UpdateBestModelFile(net, testAccuracy, parameter.Output, name, "test");
UpdateBestModelFile(net, trainAccuracy, parameter.Output, name, "train");
}
if (lr < minLearningRate)
{
Logger.Info($"Stop training: {lr} < {minLearningRate}");
break;
}
}
// wait for training threads to stop
trainer.GetNet();
Logger.Info("done training");
net.Clean();
LossMulticlassLog.Serialize(net, $"{baseName}.tmp");
// Now let's run the training images through the network. This statement runs all the
// images through it and asks the loss layer to convert the network's raw output into
// labels. In our case, these labels are the numbers between 0 and 9.
var validationParameter2 = new ValidationParameter <T, C>
{
BaseName = parameter.BaseName,
Output = parameter.Output,
Trainer = net,
TrainingImages = trainingImages,
TrainingLabels = trainingLabels,
TestingImages = testingImages,
TestingLabels = testingLabels,
UseConsole = true,
SaveToXml = true,
OutputDiffLog = true
};
Validation(validationParameter2, out _, out _);
// clean up tmp files
Clean(parameter.Output);
}
}
catch (Exception e)
{
Logger.Error(e.Message);
}
}
private static void Main(string[] args)
{
try
{
// This example is going to run on the MNIST dataset.
if (args.Length != 1)
{
Console.WriteLine("This example needs the MNIST dataset to run!");
Console.WriteLine("You can get MNIST from http://yann.lecun.com/exdb/mnist/");
Console.WriteLine("Download the 4 files that comprise the dataset, decompress them, and");
Console.WriteLine("put them in a folder. Then give that folder as input to this program.");
return;
}
Dlib.LoadMNISTDataset(args[0],
out var trainingImages,
out var trainingLabels,
out var testingImages,
out var testingLabels);
// Make an instance of our inception network.
using (var net = new LossMulticlassLog())
{
Console.WriteLine($"The net has {net.NumLayers} layers in it.");
Console.WriteLine(net);
Console.WriteLine("Training NN...");
using (var trainer = new DnnTrainer <LossMulticlassLog>(net))
{
trainer.SetLearningRate(0.01);
trainer.SetMinLearningRate(0.00001);
trainer.SetMiniBatchSize(128);
trainer.BeVerbose();
trainer.SetSynchronizationFile("inception_sync", 20);
// Train the network. This might take a few minutes...
LossMulticlassLog.Train(trainer, trainingImages, trainingLabels);
// At this point our net object should have learned how to classify MNIST images. But
// before we try it out let's save it to disk. Note that, since the trainer has been
// running images through the network, net will have a bunch of state in it related to
// the last batch of images it processed (e.g. outputs from each layer). Since we
// don't care about saving that kind of stuff to disk we can tell the network to forget
// about that kind of transient data so that our file will be smaller. We do this by
// "cleaning" the network before saving it.
net.Clean();
LossMulticlassLog.Serialize(net, "mnist_network_inception.dat");
// Now if we later wanted to recall the network from disk we can simply say:
// deserialize("mnist_network_inception.dat") >> net;
// Now let's run the training images through the network. This statement runs all the
// images through it and asks the loss layer to convert the network's raw output into
// labels. In our case, these labels are the numbers between 0 and 9.
using (var predictedLabels = net.Operator(trainingImages))
{
var numRight = 0;
var numWrong = 0;
// And then let's see if it classified them correctly.
for (var i = 0; i < trainingImages.Count; ++i)
{
if (predictedLabels[i] == trainingLabels[i])
{
++numRight;
}
else
{
++numWrong;
}
}
Console.WriteLine($"training num_right: {numRight}");
Console.WriteLine($"training num_wrong: {numWrong}");
Console.WriteLine($"training accuracy: {numRight / (double)(numRight + numWrong)}");
// Let's also see if the network can correctly classify the testing images.
// Since MNIST is an easy dataset, we should see 99% accuracy.
using (var predictedLabels2 = net.Operator(testingImages))
{
numRight = 0;
numWrong = 0;
for (var i = 0; i < testingImages.Count; ++i)
{
if (predictedLabels2[i] == testingLabels[i])
{
++numRight;
}
else
{
++numWrong;
}
}
Console.WriteLine($"testing num_right: {numRight}");
Console.WriteLine($"testing num_wrong: {numWrong}");
Console.WriteLine($"testing accuracy: {numRight / (double)(numRight + numWrong)}");
}
}
}
}
}
catch (Exception e)
{
Console.WriteLine(e);
}
}