static void Main()
{
int epochs = 6;
double learningRate = 0.01;
Optimizer gd = new GradientDescentOptimizer(learningRate);
var W = new Parameter();
var b = new Parameter();
double[] X = { 3.3, 4.4, 5.5, 6.71, 6.93, 4.168, 9.779, 6.182, 7.59, 2.167, 7.042, 10.791, 5.313, 7.997, 5.654, 9.27, 3.1 };
double[] y = { 1.7, 2.76, 2.09, 3.19, 1.694, 1.573, 3.366, 2.596, 2.53, 1.221, 2.827, 3.465, 1.65, 2.904, 2.42, 2.94, 1.3 };
for (int epoch = 0; epoch < epochs; epoch++)
{
var loss = new Variable(0);
for (int i = 0; i < X.Length; i++)
{
var prediction = W * X[i] + b;
var error = (prediction + -y[i]).Power(2);
loss += error;
}
Console.WriteLine("[Epoch: {0}] mean error = {1}", epoch, loss.Value / X.Length);
TensorSlow.Minimize(gd, loss);
}
Console.WriteLine(" y = {0}x + {1}", W.Value, b.Value);
Chart chart = new Chart();
chart.Size = new System.Drawing.Size(600, 600);
chart.ChartAreas.Add("ChartArea1");
chart.Legends.Add("legend1");
chart.Series.Add("points");
for (int i = 0; i < X.Length; i++)
{
chart.Series["points"].Points.AddXY(X[i], y[i]);
}
chart.Series["points"].ChartType = SeriesChartType.Point;
chart.Series.Add("line");
chart.Series["line"].Points.AddXY(0, b.Value);
chart.Series["line"].Points.AddXY(X.Max(), W.Value * X.Max() + b.Value);
chart.Series["line"].ChartType = SeriesChartType.Line;
chart.SaveImage("output.jpg", ChartImageFormat.Jpeg);
Console.ReadKey();
}
static void MainMNIST(string[] args)
{
Optimizer gd = new GradientDescentOptimizer(0.01);
var W = Matrix.Load("W.model");
var b = Matrix.Load("b.model");
Console.WriteLine("Loading data...");
Dataset data = LoadMNIST();
int epochs = 1, batch_size = 128;
for (int epoch = 0; epoch < epochs; epoch++)
{
for (int batch = 0; batch < 20000 / batch_size; batch++)
{
var error = new Variable(0);
for (int i = 0; i < batch_size; i++)
{
int skip = batch * batch_size + i;
Vector x = new Vector(data.Features, data.X.Skip(skip * data.Features).Take(data.Features).ToArray());
Vector y = Vector.OneHot(data.y[skip], 10);
error += Vector.ReduceSquared((W * x + b) - y) * (1.0 / batch_size);
}
TensorSlow.Minimize(gd, error);
Console.WriteLine($"[Batch {epoch + 1}.{batch + 1}] Loss: {error.Value}");
}
}
W.Save("W.model");
b.Save("b.model");
Console.ReadLine();
//var W = ReadFromBinaryFile<Matrix>("W.model");
//var b = ReadFromBinaryFile<Vector>("b.model");
int skiptest = 20000, take = 500;
double[] testX = data.X.Skip(skiptest * data.Features).Take(take * data.Features).ToArray();
int[] testY = data.y.Skip(skiptest).Take(take).ToArray();
Func <double[], int> model = x => Vector.ArgMax(W * (new Vector(data.Features, x)) + b);
Evaluate(model, testX, testY);
}