public void Train(IReadOnlyList <TrainingDataSet> trainingDataSets, double learningRate, double learningDataPercentage)
{
if (learningDataPercentage > 1 || learningDataPercentage < 0)
{
throw new ArgumentException("Percentage of training data must be a value between 0 and 1.", nameof(learningDataPercentage));
}
foreach (var dataSet in trainingDataSets)
{
if (dataSet.Inputs.Length != InputLayer.Neurons.Length)
{
throw new ArgumentException($"Data set's number of input values ({dataSet.Inputs.Length}) does not match size of input layer ({InputLayer.Neurons.Length})", nameof(trainingDataSets));
}
if (dataSet.Outputs.Length != OutputLayer.Neurons.Length)
{
throw new ArgumentException($"Data set's number of output values ({dataSet.Outputs.Length}) does not match size of output layer ({OutputLayer.Neurons.Length})", nameof(trainingDataSets));
}
}
var learningDataCount = (int)Math.Floor(trainingDataSets.Count * learningDataPercentage);
for (int i = 0; i < learningDataCount; i++)
{
InputLayer.InsertValues(trainingDataSets[i].Inputs);
HiddenLayer.Compute();
OutputLayer.Compute();
OutputLayer.ComputeDelta(trainingDataSets[i].Outputs);
HiddenLayer.ComputeDelta();
InputLayer.ComputeDelta();
HiddenLayer.UpdateWeights(learningRate);
OutputLayer.UpdateWeights(learningRate);
}
var total = 0d;
for (int i = learningDataCount; i < trainingDataSets.Count; i++)
{
var outcome = Predict(trainingDataSets[i].Inputs);
for (int j = 0; j < outcome.Length; j++)
{
total += Math.Abs(trainingDataSets[i].Outputs[j] + outcome[j]) / 2;
}
}
Accuracy = total / ((trainingDataSets.Count - learningDataCount) * OutputLayer.Neurons.Length);
}
