public void ClassifiesNAND()
{
var network = new PerceptronFactory().BuildPerceptron(3);
var trainingSet = new List<KeyValuePair<double, double[]>>
{
new KeyValuePair<double, double[]>( 1.0, new [] {1.0, 0.0, 0.0 } ),
new KeyValuePair<double, double[]>( 1.0, new [] {1.0, 0.0, 1.0 } ),
new KeyValuePair<double, double[]>( 1.0, new [] {1.0, 1.0, 0.0 } ),
new KeyValuePair<double, double[]>( 0.0, new [] {1.0, 1.0, 1.0 } ),
};
network.Train(trainingSet);
Assert.AreEqual(1.0, network.Classify(new [] { 1.0, 0.0, 0.0 }));
Assert.AreEqual(0.0, network.Classify(new [] { 1.0, 1.0, 1.0 }));
}
public void MultiLayerPerceptronClassifies()
{
// fixed arrangement for now ...
var network = new PerceptronFactory().BuildMultiLayerPerceptron();
var trainingSet = new List<KeyValuePair<double[], double[]>>
{
new KeyValuePair<double[], double[]>( new [] { 0.0, 1.0 }, new [] { 0.0, 1.0, 1.0, 0.0 } ),
new KeyValuePair<double[], double[]>( new [] { 1.0, 0.1 }, new [] { 1.0, 0.0, 1.0, 0.0 } ),
new KeyValuePair<double[], double[]>( new [] { 1.0, 1.0 }, new [] { 1.0, 1.0, 0.0, 0.0 } ),
};
network.Train(trainingSet, maxAllowedError: 0.2, maxIterations: 10000);
trainingSet.ForEach(kvp =>
{
Assert.AreEqual(kvp.Key[0], network.Classify(kvp.Value)[0]);
Assert.AreEqual(kvp.Key[1], network.Classify(kvp.Value)[1]);
});
}
public void FTClassifiesBiasedData()
{
// data has 2 double inputs and an expected binary out
var network = new PerceptronFactory().BuildFTPerceptron(2);
var reader = new TabSeparatedListReader();
List<KeyValuePair<double, double[]>> trainingSet = reader.Read(@"..\..\trainingsets\dataset2\biased.txt");
network.Train(trainingSet);
trainingSet.ForEach(kvp => Assert.AreEqual(kvp.Key, network.Classify(kvp.Value)));
}
public void SigmoidFTClassifiesNAND()
{
FTPerceptron network = new PerceptronFactory().BuildSigmoidFTPerceptron(3);
var trainingSet = new List<KeyValuePair<double, double[]>>
{
new KeyValuePair<double, double[]>( 1.0, new [] { 1.0, 0.0, 0.0 } ),
new KeyValuePair<double, double[]>( 1.0, new [] { 1.0, 0.0, 1.0 } ),
new KeyValuePair<double, double[]>( 1.0, new [] { 1.0, 1.0, 0.0 } ),
new KeyValuePair<double, double[]>( 0.0, new [] { 1.0, 1.0, 1.0 } ),
};
// gets all trained up nice well within 0.8
network.Train(trainingSet, maxAllowedError: 0.2);
// kinda hacky but basically with sigmoid activation youre never going to hit 1 and 0..
Assert.Greater(network.Classify(new[] { 1.0, 0.0, 0.0 }), 0.9);
Assert.Less(network.Classify(new[] { 1.0, 1.0, 1.0 }), 0.1);
}
public void GeneralisesSmallImages()
{
var network = new PerceptronFactory().BuildPerceptron(100);
var aSet = new InputProcessor(@"..\..\trainingsets\dataset1\A");
var bSet = new InputProcessor(@"..\..\trainingsets\dataset1\B");
var trainingSet = new List<KeyValuePair<double, double[]>>
{
new KeyValuePair<double, double[]>( 1.0, aSet.BmpToBinaryArr(@"1.jpg") ),
new KeyValuePair<double, double[]>( 1.0, aSet.BmpToBinaryArr(@"2.jpg") ),
new KeyValuePair<double, double[]>( 1.0, aSet.BmpToBinaryArr(@"3.jpg") ),
new KeyValuePair<double, double[]>( 1.0, aSet.BmpToBinaryArr(@"4.jpg") ),
new KeyValuePair<double, double[]>( 1.0, aSet.BmpToBinaryArr(@"5.jpg") ),
new KeyValuePair<double, double[]>( 0.0, bSet.BmpToBinaryArr(@"1.jpg") ),
new KeyValuePair<double, double[]>( 0.0, bSet.BmpToBinaryArr(@"2.jpg") ),
new KeyValuePair<double, double[]>( 0.0, bSet.BmpToBinaryArr(@"3.jpg") ),
new KeyValuePair<double, double[]>( 0.0, bSet.BmpToBinaryArr(@"4.jpg") ),
new KeyValuePair<double, double[]>( 0.0, bSet.BmpToBinaryArr(@"5.jpg") ),
};
network.Train(trainingSet);
Assert.AreEqual(1.0, network.Classify(aSet.BmpToBinaryArr(@"7.jpg")));
Assert.AreEqual(0.0, network.Classify(bSet.BmpToBinaryArr(@"7.jpg")));
}