Na⁺ - A Powerful Neural Network Library for .NET

Training

If you are installing with Nuget it is necessary to install both SodiumPlus and SodiumPlusTraining to run training.

Example of training using error backpropagation to solve an XOr problem

var trainingPatterns = new List<TrainingPattern>
{
    new TrainingPattern(new [] { 0d, 0d }, new [] { 0d, 1d }), // 0d, 1d == false
    new TrainingPattern(new [] { 1d, 0d }, new [] { 1d, 0d }), // 1d, 0d == true
    new TrainingPattern(new [] { 0d, 1d }, new [] { 1d, 0d }), // 1d, 0d == true
    new TrainingPattern(new [] { 1d, 1d }, new [] { 0d, 1d }), // 0d, 1d == false
};

var perceptron = await new ErrorBackPropagationBuilder()
    .With.ANewLayerOfInputUnits(2)
    .ConnectedTo.ANewLayerOfHiddenUnits(3).With.SigmoidActivation()
    .ConnectedTo.ANewLayerOfOutputUnits(2).With.SoftmaxActivation()
    .And.UseCrossEntropyErrorFunction()
    .And.Bias(0.5)
    .And.LearningRate(1d)
    .And.Momentum(0d)
    .And.SlopeMultiplier(1d)
    .And.UseOneHotEncoding()
    .And.SetupNetwork()
    .And.NameEverything()
    .And.ReadyForTraining()
    .TrainAsync(trainingPatterns, 0.1d, 1000);

Classification

Now use the perceptron to classify inputs

    var result1 = await perceptron.FireAsync(new [] { 0d, 0d });
    Console.WriteLine("(0, 0) -> (" + string.Join(",", result1) + ")");

    var result2 = await perceptron.FireAsync(new[] { 1d, 0d });
    Console.WriteLine("(1, 0) -> (" + string.Join(",", result2) + ")");

    var result3 = await perceptron.FireAsync(new[] { 0d, 1d });
    Console.WriteLine("(0, 1) -> (" + string.Join(",", result3) + ")");

    var result4 = await perceptron.FireAsync(new[] { 1d, 1d });
    Console.WriteLine("(1, 1) -> (" + string.Join(",", result4) + ")");

Saving/Loading

Store the perceptron for later use using the Serialization library...

    File.WriteAllText("perceptron.txt", new PerceptronSerializer().SerializeJson(perceptron));

...and then reload it...

    var perceptron = new PerceptronSerializer().DeserializeJson(File.ReadAllText("perceptron.txt"));