How exactly one goes about learning XOR in with this tool?

[mvrozanti]

I don't get the concept of 'classes' of the README example.

I know the purpose of this tool is to deal with multidimensional input but I'm a bit stuck

[elamaunt]

Hi!
I tried to implement same code with convnetjs from https://gist.github.com/willguitaradmfar/12f7efb2ab9534755619ba4dce72f988

Js sample works fine:

But same code on sharp doesnt give same result:

        var net = new Net<double>();
        
        net.AddLayer(new InputLayer(1, 1, 2));

        net.AddLayer(new FullyConnLayer(3));
        net.AddLayer(new TanhLayer());

        net.AddLayer(new FullyConnLayer(2));
        net.AddLayer(new SoftmaxLayer(2));


        var x0 = new Volume(new double[] { 1, 1 }, new Shape(2));
        var x1 = new Volume(new double[] { 0, 1 }, new Shape(2));
        var x2 = new Volume(new double[] { 1, 0 }, new Shape(2));
        var x3 = new Volume(new double[] { 0, 0 }, new Shape(2));
        
        Console.WriteLine("x0: " + net.Forward(x0).Get(0));
        Console.WriteLine("x1: " + net.Forward(x1).Get(0));
        Console.WriteLine("x2: " + net.Forward(x2).Get(0));
        Console.WriteLine("x3: " + net.Forward(x3).Get(0));

        Console.WriteLine();
        Console.WriteLine();
        
        var trainer = new SgdTrainer<double>(net) { LearningRate = 0.2, L2Decay = 0.001 };
        
        for (int i = 0; i < 2000; i++)
        {
            trainer.Train(x0, new Volume(new double[] { 0.0 }, new Shape(1)));
            trainer.Train(x1, new Volume(new double[] { 1.0 }, new Shape(1)));
            trainer.Train(x2, new Volume(new double[] { 1.0 }, new Shape(1)));
            trainer.Train(x3, new Volume(new double[] { 0.0 }, new Shape(1)));
        }

        Console.WriteLine("x0: " + net.Forward(x0).Get(0));
        Console.WriteLine("x1: " + net.Forward(x1).Get(0));
        Console.WriteLine("x2: " + net.Forward(x2).Get(0));
        Console.WriteLine("x3: " + net.Forward(x3).Get(0));

Perhaps there are some errors in my code or in the project code.

[elamaunt]

Changing of shape changes results. I could not understand how does shapes work =/

[mvrozanti]

I've given up on C# approaches. Python is much more readable and while it may not be as 'debuggable' it is easier to understand where you're screwing up. Also so many good libraries.

[cbovar]

From issue #74

var net = new Net<double>();
net.AddLayer(new InputLayer(1, 1, 2)); // Two inputs
net.AddLayer(new FullyConnLayer(3));
net.AddLayer(new TanhLayer());
net.AddLayer(new FullyConnLayer(2));
net.AddLayer(new SoftmaxLayer(2)); // Two possible outcome/class (0 and 1)

// All possible inputs
var x0 = new Volume(new double[] { 1, 1 }, new Shape(2));
var x1 = new Volume(new double[] { 0, 1 }, new Shape(2));
var x2 = new Volume(new double[] { 1, 0 }, new Shape(2));
var x3 = new Volume(new double[] { 0, 0 }, new Shape(2));

var trainer = new SgdTrainer<double>(net) { LearningRate = 0.2 };

for (var i = 0; i < 2000; i++)
{
    trainer.Train(x0, new Volume(new[] { 1.0, 0.0 }, new Shape(1, 1, 2, 1)));
    trainer.Train(x1, new Volume(new[] { 0.0, 1.0 }, new Shape(1, 1, 2, 1)));
    trainer.Train(x2, new Volume(new[] { 0.0, 1.0 }, new Shape(1, 1, 2, 1)));
    trainer.Train(x3, new Volume(new[] { 1.0, 0.0 }, new Shape(1, 1, 2, 1)));

    Console.WriteLine(trainer.Loss); // Should decrease
}

Console.WriteLine("x0: " + net.Forward(x0).Get(0));
Console.WriteLine("x1: " + net.Forward(x1).Get(0));
Console.WriteLine("x2: " + net.Forward(x2).Get(0));
Console.WriteLine("x3: " + net.Forward(x3).Get(0));

Console.ReadLine();