NeuralNetLib

This project implements a basic neural network and training algorithms.

Cake Tasks

This project uses Cake!

• ./cake.ps1 -target=Clean
• ./cake.ps1 -target=Restore-Nuget-Packages
• ./cake.ps1 -target=Build
• ./cake.ps1 -target=Test

CI

	Windows	Linux
Master

Example

Neural Net Lib supports back propagation:

Net net = new Net(new Random(), 2, 1);

BackPropagation prop = new BackPropagation(2, 1);
DataSet _1 = new DataSet(new double[] { 0, 0 }, new double[] { 0 });    // 0 | 0 = 0
DataSet _2 = new DataSet(new double[] { 0, 1 }, new double[] { 1 });    // 0 | 1 = 1
DataSet _3 = new DataSet(new double[] { 1, 0 }, new double[] { 1 });    // 1 | 0 = 1
DataSet _4 = new DataSet(new double[] { 1, 1 }, new double[] { 0 });    // 1 | 1 = 0

prop.AddDataSet(_1);
prop.AddDataSet(_2);
prop.AddDataSet(_3);
prop.AddDataSet(_4);

int epoch = 1000;
var backPropResult = prop.Train(net, epoch);

Assert.IsTrue(backPropResult.SSE < 0.2, "LogicNetXOR SSE after {0} epochs is '{1}'", epoch, backPropResult.SSE);

Genetic training is also a thing:

int iterations = 1000;
int population = 100;

DataSet _1 = new DataSet(new double[] { 0, 0 }, new double[] { 0 });    // 0 | 0 = 0
DataSet _2 = new DataSet(new double[] { 0, 1 }, new double[] { 1 });    // 0 | 1 = 1
DataSet _3 = new DataSet(new double[] { 1, 0 }, new double[] { 1 });    // 1 | 0 = 1
DataSet _4 = new DataSet(new double[] { 1, 1 }, new double[] { 0 });    // 1 | 1 = 0

var dataSets = new[] { _1, _2, _3, _4 };

var fitnessEvaluator = new DatasetEvaluator(dataSets);
var trainer = new GeneticAlgorithmTrainer<DatasetEvaluator>(new Random(), fitnessEvaluator);

var topNet = trainer.TrainAi(_1.InputCount, _1.OutputCount, 3, population, iterations).First();

double sse = 0;
foreach (var dataSet in dataSets)
{
    double result = topNet.Calculate(dataSet.Inputs).First();
    sse += Math.Pow(dataSet.Outputs.First() - result, 2.0);
}

Assert.IsTrue(sse < 0.2, $"LogicNetXOR SSE after {iterations} iterations is '{sse}'");