public virtual void Train(int epochs, int numSamples, int batchSize,
LearningOptions learningOptions,
CallBackOptions callBackOptions = default)
{
LearningBegan?.Invoke(this, new LearningEventArgs());
for (int epoch = 0; epoch < epochs; epoch++)
{
for (int i = 0; i < (numSamples / batchSize); i++)
{
ProcessLearning(Data.ReadNext(new StreamOptions(batchSize, learningOptions.shuffleData)),
learningOptions.learningRate, learningOptions.momentumRate, callBackOptions);
}
if (numSamples % batchSize > 0)
{
ProcessLearning(Data.ReadNext(new StreamOptions(numSamples % batchSize, learningOptions.shuffleData)),
learningOptions.learningRate, learningOptions.momentumRate, callBackOptions);
}
if (callBackOptions.call == CallTime.EveryEpoch)
{
LearningPerformed?.Invoke(this, new LearningEventArgs());
}
}
LearningEnded?.Invoke(this, new LearningEventArgs());
}
private void ProcessLearning(Sample[] samples, float eta, float alpha, CallBackOptions callBackOptions)
{
foreach (Sample sample in samples)
{
Net.ProcessInputs(sample.inputs);
Net.Learn(sample.outputs, eta, alpha);
if (callBackOptions.call == CallTime.EverySample)
{
Cost(sample.outputs);
LearningPerformed?.Invoke(this, new LearningEventArgs(sample, Net.Prediction, MSE));
}
}
if (callBackOptions.call == CallTime.EveryBatch)
{
Cost(samples[samples.Length - 1].outputs);
LearningPerformed?.Invoke(this, new LearningEventArgs(samples[samples.Length - 1], Net.Prediction, MSE));
}
}
