public void TestNetworkWorkingAtAll_TryAndOperation()
{
IActivationFunction func = new BipolarSigmoid(2.0);
var layerCounts = new List<int>();
layerCounts.Add(2);
layerCounts.Add(2);
layerCounts.Add(1);
var network = new Network(2, 3, layerCounts, func, new RandomInitializer());
//double[] input = new double[] { -3, -2, -1, 0, 1, 2, 3 };
//double[] output = new double[] { 9, 4, 1, 0, 1, 4, 9 };
double[][] input = new double[][] { new[] { 0.0, 0.0 }, new[] { 1.0, 0.0 }, new[] { 0.0, 1.0 }, new[] { 1.0, 1.0 } };
double[] output = new double[] { 0.0, 1.0, 1.0, 1.0 };
//double[] input = new double[] { 1,2,3,4,5,6,7,8,9 };
//double[] output = new double[]{ 0,4,7,9,10,9,7,4,0 };
var learner = new Teacher(network);
learner.Rate = 0.2;
learner.Momentum = 0.0;
List<double[]> netInputs = new List<double[]>();
List<double[]> netOutputs = new List<double[]>();
for (int i = 0; i < input.Length; i++)
{
netInputs.Add(input[i]);
netOutputs.Add(new double[]{output[i]});
}
for (int i = 0; i < 1000; i++)
{
learner.TeachOnSamples(netInputs, netOutputs);
}
double[] aproximated = new double[input.Length];
Console.WriteLine("Try those Values in Excel or whatever");
using (var writer = new StreamWriter("data_dump.txt"))
{
for (int i = 0; i < aproximated.Length; i++)
{
aproximated[i] = network.ComputeOutputVector(input[i])[0];
writer.WriteLine((aproximated[i] ).ToString());
}
}
}
public Teacher(Network net)
{
this.net = net;
errors = new double[net.LayersCount][];
updates = new double[net.LayersCount][][];
noise = new double[net.LayersCount][];
for (int i = 0; i < net.LayersCount; i++)
{
errors[i] = new double[net[i].Size];
noise[i] = new double[net[i].Size];
updates[i] = new double[net[i].Size][];
for (int j = 0; j < net[i].Size; j++)
{
updates[i][j] = new double[net[i].Inputs];
}
}
}
public void TestTimeSeriesPrediction()
{
const int windowSize = 5;
const int predictionSize = 1;
IActivationFunction func = new BipolarSigmoid(2.0);
var layerCounts = new List<int>();
layerCounts.Add(windowSize);
layerCounts.Add(2*windowSize);
layerCounts.Add(1);
var network = new Network(5, 3, layerCounts, func, new RandomInitializer());
//double[] input = new double[] { -3, -2, -1, 0, 1, 2, 3 };
//double[] output = new double[] { 9, 4, 1, 0, 1, 4, 9 };
double[] readedData = new double[50];
double[] data;
double min, max;
double range;
using(var reader = new StreamReader(File.OpenRead("sinusoid.csv")))
{
int i = 0;
readedData[i] = double.Parse(reader.ReadLine(), System.Globalization.CultureInfo.InvariantCulture);
min = readedData[i];
max = readedData[i];
try
{
for (i = 1; i < 50; i++)
{
readedData[i] = double.Parse(reader.ReadLine(), System.Globalization.CultureInfo.InvariantCulture);
if (min > readedData[i])
min = readedData[i];
if (max < readedData[i])
max = readedData[i];
}
}
catch (Exception)
{
}
if (i > 0)
{
data = new double[i];
Array.Copy(readedData, data, i);
}
else return;
}
range = (max - min);
int learningSamples = data.Length - windowSize - predictionSize;
double[][] input = new double[learningSamples][];
double[][] output = new double[learningSamples][];
for (int i = 0; i < learningSamples; i++)
{
input[i] = new double[windowSize];
output[i] = new double[1];
for (int j = 0; j < windowSize; j++)
{
input[i][j] = TransformData(data[i + j], min, range);
}
output[i][0] = TransformData(data[i + windowSize], min, range);
}
//double[] input = new double[] { 1,2,3,4,5,6,7,8,9 };
//double[] output = new double[]{ 0,4,7,9,10,9,7,4,0 };
var learner = new Teacher(network);
learner.Rate = 0.1;
learner.Momentum = 0.0;
var solution = new double[data.Length - windowSize, 2];
var netInput = new double[windowSize];
for (int i = 0; i < 1000; i++)
{
var ins = new List<double[]>();
var outs = new List<double[]>();
for (int ii = 0; ii < input.Length; ii++)
{
ins.Add(input[ii]);
outs.Add(output[ii]);
}
learner.TeachOnSamples(ins, outs);
for (int j = 0; j < data.Length - windowSize; j++)
{
for (int k = 0; k < windowSize; k++)
{
netInput[k] = TransformData(data[j + k], min, range) ;
}
solution[j,1]= TransformBack(network.ComputeOutputVector(netInput)[0], min, max );//(network.ComputeOutputVector(netInput)[0]) / range + min;
}
}
double[] aproximated = new double[data.Length - windowSize];
Console.WriteLine("Try those Values in Excel or whatever");
using (var writer = new StreamWriter("data_dump_time_series.txt"))
{
for (int i = 0; i < aproximated.Length; i++)
{
aproximated[i] = solution[i,1];
writer.WriteLine((aproximated[i]).ToString());
}
}
}
public void TestXMLSerializationAndDeserialization()
{
IActivationFunction func = new BipolarSigmoid(2.0);
var layerCounts = new List<int>();
layerCounts.Add(5);
layerCounts.Add(10);
layerCounts.Add(1);
var network = new Network(5, 3, layerCounts, func, new RandomInitializer());
var serializer = new DataContractSerializer(typeof(Network), new[] { func.GetType(), typeof(RandomInitializer) });
var settings = new XmlWriterSettings();
settings.Indent = true;
settings.IndentChars = " ";//4 spaces
string file = "serialized_net.xml";
using (var writer = XmlWriter.Create(file, settings))
{
serializer.WriteObject(writer, network);
}
Network newNet = (Network)serializer.ReadObject(XmlReader.Create(file));
Assert.IsNotNull(newNet);
Assert.AreEqual(5, newNet.Inputs);
Assert.AreEqual(3, newNet.LayersCount);
Assert.IsTrue(0 < newNet[0][0].Weights[0]);
Assert.AreEqual(typeof(BipolarSigmoid), newNet[0][0].ActivationFunction.GetType());
}
public void TestXMLSerialization()
{
IActivationFunction func = new BipolarSigmoid(2.0);
var layerCounts = new List<int>();
layerCounts.Add(5);
layerCounts.Add(10);
layerCounts.Add(1);
var network = new Network(5, 3, layerCounts, func, new RandomInitializer());
var serializer = new DataContractSerializer(typeof(Network),new[]{func.GetType(), typeof(RandomInitializer)});
var fileName = "serialized_net.xml";
var settings = new XmlWriterSettings();
settings.Indent = true;
settings.IndentChars = " ";//4 spaces
using (var writer = XmlWriter.Create(fileName, settings))
{
serializer.WriteObject(writer, network);
}
var fInfo = new FileInfo(fileName);
Assert.IsTrue(fInfo.Exists);
Assert.IsTrue(fInfo.Length > 0);
}
public void SetSettings()
{
activ = AISettings.Activ;
init = AISettings.Init;
layer = AISettings.Layer;
network = new Network(layer[0], layer.Count, layer, activ, init);
learner = new Teacher(network);
rate = AISettings.LearnerRate;
momentum = AISettings.LearnerMomentum;
eraCount = AISettings.IterationsCount;
indicator = AISettings.Indicator;
}