public void setInput(EMNistDecoder _emnistDecoder, int imageNumber)
{
int k = 0;
for (int i = 0; i < _emnistDecoder.numRows; i++)
{
for (int j = 0; j < _emnistDecoder.numCols; j++)
{
inputArray[k] = _emnistDecoder.images[imageNumber, i, j] == 0 ? 0 : 1;
k++;
}
}
}
public Input(EMNistDecoder _emnistDecoder)
{
inputArray = new double[_emnistDecoder.dimensions];
}
static void Main()
{
int layerSize;
int inputLayerSize;
int outputLayerSize;
int networkSizeCounter = 0; //Counts Network Size for display
int connectionSizeCounter = 0; //Calculates Connection Size for Display
int imageAndLabelCounter = 0;
float imageCount;
float succesfulPredictions = 0;
float unsuccesfulPredictions = 0;
int[] layerAndNodeAmountHolder;
string labelFilePath;
string imagesFilePath;
labelFilePath = @"C:\EMNist Dataset\gzip\gzip\emnist-mnist-train-labels-idx1-ubyte\emnist-mnist-train-labels-idx1-ubyte";
imagesFilePath = @"C:\EMNist Dataset\gzip\gzip\emnist-mnist-train-images-idx3-ubyte\emnist-mnist-train-images-idx3-ubyte";
/*
* Console.WriteLine("Please enter labelFilePath");
* while(true)
* {
* labelFilePath = Console.ReadLine();
* if(File.Exists(labelFilePath))
* {
* break;
* }
* else
* {
* Console.WriteLine("Entered wrong file location. Please enter labelFilePath");
* }
* }
*
* Console.WriteLine("Please enter imagesFilePath");
*
* while (true)
* {
* imagesFilePath = Console.ReadLine();
* if (File.Exists(imagesFilePath))
* {
* break;
* }
* else
* {
* Console.WriteLine("Entered wrong file location. Please enter imagesFilePath");
* }
* }
*/
EMNistDecoder emnistDecoder = new EMNistDecoder();
emnistDecoder.EMNistDecoderInit(labelFilePath, imagesFilePath);
imageCount = emnistDecoder.numImages;
inputLayerSize = emnistDecoder.dimensions;
Console.WriteLine("Please enter classification output parameter");
outputLayerSize = Convert.ToInt32(Console.ReadLine());
Variables.OutputSize = outputLayerSize;
Variables.LearningRate = 0.05f;
Output outputArray = new Output(Variables.OutputSize);
Stopwatch stopwatch = new Stopwatch();
Network network = new Network();
//Variables variables = new Variables();
Random rand = new Random();
Console.WriteLine("Please Enter Hidden Layer Amount");
Variables.LayerAmount = Convert.ToInt32(Console.ReadLine());
layerAndNodeAmountHolder = new int[Variables.LayerAmount];
//Console.WriteLine("Please enter Input Layer Size");
//layerSize = Convert.ToInt32(Console.ReadLine());
//layerAndNodeAmountHolder[0] = layerSize;
Variables.InputSize = inputLayerSize;
Input inputArray = new Input(emnistDecoder);
for (int i = 0; i < layerAndNodeAmountHolder.Length; i++)
{
Console.WriteLine("Please enter " + (i + 1) + "th Hidden Layer Size");
layerSize = Convert.ToInt32(Console.ReadLine());
layerAndNodeAmountHolder[i] = layerSize;
}
//Console.WriteLine("Please enter Output Layer Size");
//layerSize = Convert.ToInt32(Console.ReadLine());
// layerSize = emnistDecoder.labels.Length;
//layerAndNodeAmountHolder[Variables.LayerAmount - 1] = layerSize;
stopwatch.Start();
network.addLayerToNetwork(inputLayerSize, 0);
for (int i = 0; i < layerAndNodeAmountHolder.Length; i++)
{
network.addLayerToNetwork(layerAndNodeAmountHolder[i], 1);
}
network.addLayerToNetwork(outputLayerSize, 2);
networkSizeCounter = network.networkSize;
for (int i = 0; i < network.networkSize - 1; i++)
{
connectionSizeCounter += network.Layers[i].layerSize * network.Layers[i + 1].layerSize;
}
stopwatch.Stop();
Console.WriteLine("Network has " + networkSizeCounter + " Layers");
Console.WriteLine("Network has " + NodeCounter + " Neurons");
Console.WriteLine("Network has " + connectionSizeCounter + " Connections");
Console.WriteLine("Creation of the network took " + stopwatch.Elapsed);
//Console.WriteLine("Please enter Training Data Set Locaiton");
//variables.TrainingFilePath = Console.ReadLine();
Console.WriteLine("Magic number of training set is =" + emnistDecoder.magic1);
Console.WriteLine("Image Count in database is = " + emnistDecoder.numImages);
Console.Read();
Console.Clear();
imageAndLabelCounter = 0;
while (imageAndLabelCounter < imageCount)
{
//emnistDecoder.emnistDecoderPrint(imageAndLabelCounter);
inputArray.setInput(emnistDecoder, imageAndLabelCounter);
//inputArray.debugPrintInput();
//Thread.Sleep(1000);
outputArray.setExpectedOutputArray(emnistDecoder.getCurrentImageLabel(imageAndLabelCounter));
network.setInputLayerInputs(inputArray);
network.feedForward(network);
//network.printNetworkLayerOutputDelta(2);
//network.printNetworkLayerBias(2);
//network.printNetworkLayerOutput(2);
Console.WriteLine("Network prediction was : " + network.getPrediction());
//Console.ReadLine();
if (network.getPrediction() == emnistDecoder.getCurrentImageLabel(imageAndLabelCounter))
{
Console.WriteLine("Succesful Prediction!");
//Console.WriteLine("\n##Succesful Values##");
//network.printNetworkLayerOutputDelta(2);
//network.printNetworkLayerBias(2);
//network.printNetworkLayerOutput(2);
succesfulPredictions++;
}
else
{
Console.WriteLine("Bad Prediction :(...");
//Console.WriteLine("##Before Values##");
//network.printNetworkLayerOutputDelta(2);
//network.printNetworkLayerBias(2);
//network.printNetworkLayerOutput(2);
unsuccesfulPredictions++;
network.calculateOutputError(outputArray);
network.startBackPropogation();
network.updateWeightsAndBiases();
//Console.WriteLine("\n##After Values##");
//network.printNetworkLayerOutputDelta(2);
//network.printNetworkLayerBias(2);
//network.printNetworkLayerOutput(2);
}
Console.WriteLine("Success Rate is : %" + ((succesfulPredictions / (imageAndLabelCounter + 1)) * 100));
Console.WriteLine("Training Progress : " + ((imageAndLabelCounter + 1) / imageCount) * 100);
imageAndLabelCounter++;
network.printNetworkLayerOutputDelta(2);
network.printNetworkLayerBias(2);
network.printNetworkLayerOutput(2);
//network.printInputDebug();
//Console.ReadLine();
outputArray.resetExpectedOutputArray();
network.resetAllDeltaValues();
Console.SetCursorPosition(0, 0);
// Console.ReadLine();
// Console.Clear();
// Thread.Sleep(1000);
}
/* for (int i = 0; i < 1000; i++)
* {
* mnistDecoder.mNistSetImage();
* mnistDecoder.mNistSetLabel();
*
* Console.WriteLine();
* }
*/
Console.ReadLine();
}