public void Train(string imagesPath, string labelsPath, double learningRate)
{
foreach (var image in MnistReader.ReadData(imagesPath, labelsPath))
{
int[] desiredOutputs = new int[10];
Byte[] inps = image.Data.Cast <Byte>().ToArray();
List <double> predictions = new List <double>();
List <double> inputs = inps.Select(i => (double)i / 255.0).ToList();
//Console.WriteLine((int)image.Label);
desiredOutputs[image.Label] = 1;
predictions = ForwardPass(inputs);
BackwardPass(predictions, desiredOutputs, learningRate);
//Console.WriteLine("Desired outputs ");
//foreach (var elt in desiredOutputs)
//{
// Console.Write(" " + elt);
//}
//Console.WriteLine();
//predictions = ForwardPass(inputs);
//Console.WriteLine("Predictions");
//foreach (var elt in predictions)
//{
// Console.Write(" " + elt);
//}
//BackwardPass(predictions, desiredOutputs);
}
}
public Tuple <double, double> evaluate(string imagesPath, string labelsPath)
{
var accuracyValue = 0;
var numberImage = 0;
double loss = 0.0;
foreach (var image in MnistReader.ReadData(imagesPath, labelsPath))
{
int[] desiredOuput = new int[10];
desiredOuput[image.Label] = 1;
numberImage += 1;
Byte[] inps = image.Data.Cast <Byte>().ToArray();
List <double> predictions = new List <double>();
List <double> input = inps.Select(i => (double)i / 255.0).ToList();
predictions = ForwardPass(input);
//foreach (var elt in predictions)
//{
// Console.Write(" {0}", elt);
//}
//Console.WriteLine();
var predictionValue = predictions.IndexOf(predictions.Max());
var maxValue = predictions.Max();
//Console.WriteLine("Expected {0} : predicted {1}: max value {2} ", image.Label, predictionValue, maxValue);
if (predictionValue == image.Label)
{
accuracyValue += 1;
}
loss += Utils.calculate_loss(predictions, desiredOuput);
}
double averageLoss = loss / numberImage;
double averageAccuracy = (double)accuracyValue / numberImage;
return(Tuple.Create(averageLoss, averageAccuracy));
}
