static void Main(string[] args)
{
//var net = MNIST_Train();
//net.SaveModel("MNIST_model.json");
//net.SaveModel("../MNIST_model.json");
var net = new SimpleNeuralNet(784, 200, 10);
net.Initialize();
net.LoadModel("../MNIST_model.json");
Kaggle_MNIST_Test(net);
}
private static void Kaggle_MNIST_Test(SimpleNeuralNet net)
{
Console.WriteLine("Testing...");
string fname = "../kaggle_mnist_test.csv";
//string fname = "../my_test_data.csv";
//string fname = "../mnist_train.csv";
var lines = File.ReadAllLines(fname);
var train_data = new Vector <double> [lines.Length - 1];
var j = -1;
foreach (var line in lines)
{
if (line.Contains("pixel"))
{
continue;
}
if (string.IsNullOrEmpty(line))
{
continue;
}
var words = line.Split(',');
string label = words[0];
double[] nums = new double[words.Length];
for (int i = 0; i < words.Length; i++)
{
nums[i] = Convert.ToDouble(words[i]);
nums[i] = (nums[i] / 255.0) * 0.99 + 0.01; // normalize the nums to be between 0.01 - 1 (inclusive)
}
train_data[++j] = DenseVector.OfArray(nums);
}
lines = null;
int right = 0;
string fileContents = "ImageId,Label\n";
for (int i = 0; i < train_data.Length; i++)
{
var answer = net.Query(train_data[i]);
int max = 0;
for (int k = 1; k < answer.Count; k++)
{
if (answer[k] > answer[max])
{
max = k;
}
}
fileContents += (i + 1) + "," + max + "\n";
}
File.WriteAllText("../kaggle_mnist_answer.csv", fileContents);
}
private static SimpleNeuralNet MNIST_Train(int epochs = 10)
{
var lines = File.ReadAllLines("../mnist_train.csv");
Vector <double>[] train_data = new Vector <double> [lines.Length];
Vector <double>[] target_data = new Vector <double> [lines.Length];
var j = -1;
foreach (var line in lines)
{
var words = line.Split(',');
string label = words[0];
double[] nums = new double[words.Length - 1];
for (int i = 1; i < words.Length; i++)
{
nums[i - 1] = Convert.ToDouble(words[i]);
nums[i - 1] = (nums[i - 1] / 255.0) * 0.99 + 0.01; // normalize the nums to be between 0.01 - 1 (inclusive)
}
train_data[++j] = DenseVector.OfArray(nums);
var target = new double[10];
for (int i = 0; i < 10; i++)
{
target[i] = 0.01;
}
target[Convert.ToInt32(label)] = 0.99;
target_data[j] = DenseVector.OfArray(target);
}
Console.WriteLine("Initialize the network...");
var net = new SimpleNeuralNet(train_data[0].Count, 200, 10);
net.Initialize();
for (int k = 0; k < epochs; k++)
{
Console.WriteLine("start training: epoch[" + (k + 1) + "]");
for (int i = 0; i < train_data.Length; i++)
{
net.Train(train_data[i], target_data[i]);
}
}
return(net);
}
private static void MNIST_Test(SimpleNeuralNet net)
{
Console.WriteLine("Testing...");
string fname = "../mnist_test.csv";
//string fname = "../my_test_data.csv";
//string fname = "../mnist_train.csv";
var lines = File.ReadAllLines(fname);
var train_data = new Vector <double> [lines.Length];
var target_data = new Vector <double> [lines.Length];
var j = -1;
foreach (var line in lines)
{
var words = line.Split(',');
string label = words[0];
double[] nums = new double[words.Length - 1];
for (int i = 1; i < words.Length; i++)
{
nums[i - 1] = Convert.ToDouble(words[i]);
nums[i - 1] = (nums[i - 1] / 255.0) * 0.99 + 0.01; // normalize the nums to be between 0.01 - 1 (inclusive)
}
train_data[++j] = DenseVector.OfArray(nums);
var target = new double[10];
for (int i = 0; i < 10; i++)
{
target[i] = 0.01;
}
target[Convert.ToInt32(label)] = 0.99;
target_data[j] = DenseVector.OfArray(target);
}
lines = null;
int right = 0;
for (int i = 0; i < train_data.Length; i++)
{
Console.WriteLine("Test #" + (i + 1));
int right_answer = 0;
for (int k = 0; k < target_data[i].Count; k++)
{
if (target_data[i][k] > 0.98)
{
right_answer = k;
break;
}
}
var answer = net.Query(train_data[i]);
int max = 0;
for (int k = 1; k < answer.Count; k++)
{
if (answer[k] > answer[max])
{
max = k;
}
}
right += (max == right_answer ? 1 : 0);
Console.WriteLine(right_answer + " NI: " + max + " " + (max == right_answer ? "R" : "W"));
}
Console.WriteLine("Accuracy: " + ((double)right / (double)train_data.Length));
}