public void Train(MnistImageReader reader)
{
// Randomize network's parameters
RandomizeWeights();
RandomizeBiases();
var batchSize = 1;
var iterations = reader.ImagesCount / batchSize;
var matches = 0;
Console.WriteLine("Training started");
var stopwatch = new Stopwatch();
stopwatch.Restart();
for (int i = 0; i < iterations; i++)
{
if (TrainBatch(batchSize, i, reader))
{
matches++;
}
Console.Clear();
Console.WriteLine($"{(double)(i+1)/iterations*100:F}%");
}
stopwatch.Stop();
var elapsedTime = stopwatch.Elapsed;
Console.WriteLine($"Trained on {reader.ImagesCount} images in {elapsedTime.Minutes}:{elapsedTime.Seconds}.{elapsedTime.Milliseconds}s.");
}
private bool TrainBatch(int batchSize, int iteration, MnistImageReader reader)
{
int i = 0, matches = 0;
while (i < batchSize)
{
var(image, label) = reader.ReadImage();
// Upload image into input layer
UploadImage(image);
// Create expected results vector
var expectedResults = new double[activations[^ 1].Length];
public int Recognize(MnistImageReader reader, int numberOfImages)
{
int matches = 0;
for (int i = 0; i < numberOfImages; i++)
{
var(image, label) = reader.ReadImage();
var result = Recognize(image);
var match = result == label;
Console.WriteLine($"{i+1}: match={match}, in={label}, out={result}");
if (match)
{
matches++;
}
}
Console.WriteLine($"RESULT: {((double)matches/(double)numberOfImages*100.0):F}% ({matches}/{numberOfImages})");
return(numberOfImages);
}
