public Network(int[] neuronCounts, IFunction activator, IDataset dataset, double dataRatio, double?weight = null)
{
// throw some errors first
if (neuronCounts.Length < 2)
{
throw new ArgumentOutOfRangeException("Network must have at least 2 layers");
}
if (neuronCounts[0] != dataset.GetDataSize())
{
throw new ArgumentOutOfRangeException("Input neuron count must match 1:1 with input data size. Current neuron count: " + neuronCounts[0] + ", current input size: " + dataset.GetDataSize());
}
if (neuronCounts[neuronCounts.Length - 1] != dataset.GetLabelSize())
{
throw new ArgumentOutOfRangeException("Output neuron count must match 1:1 with output data size. Current neuron count: " + neuronCounts[neuronCounts.Length - 1] + ", current input size: " + dataset.GetLabelSize());
}
// init new network based on neuroncounts
Layers = new Layer[neuronCounts.Length];
Layers[0] = new Layer(neuronCounts[0]);
for (int c = 1; c < neuronCounts.Length; c++)
{
Layers[c] = new Layer(neuronCounts[c], Layers[c - 1], activator, weight);
}
// create testdata and trainingdata
double counter = Math.Round(1 / (1 - dataRatio)); // if data ratio is 0.8 then 80% of the data will be for training.
// this means the counter should be 5, eg. every 5th entry will be test data rather than training data
double[][] inputSet = dataset.GetDataset();
double[][] outputSet = dataset.GetLabelset();
List <double[]> tr_i = new List <double[]> {
};
List <double[]> tr_o = new List <double[]> {
};
List <double[]> te_i = new List <double[]> {
};
List <double[]> te_o = new List <double[]> {
};
for (int i = 0; i < inputSet.Length; i++)
{
if (i % counter == 0) // how we test whether to add something to test data or training data
{
// test data goes in here
te_i.Add(inputSet[i]);
te_o.Add(outputSet[i]);
}
else
{
// training data goes in here
tr_i.Add(inputSet[i]);
tr_o.Add(outputSet[i]);
}
}
TrainingInputs = tr_i.ToArray();
TrainingOutputs = tr_o.ToArray();
TestingInputs = te_i.ToArray();
TestingOutputs = te_o.ToArray();
GlobalError = 1337;
}