void BatchGD(T[] inputData, T[] knownOutputs)
{
List <double> wts = new List <double>(new double[Weights.Count]);
Parallel.For(0, inputData.Length, (i) =>
{
Parallel.For(0, Weights.Count, (j) =>
{
double delWT = ((dynamic)knownOutputs[i] - Predict(inputData[i])) * System.Math.Pow((double)(dynamic)inputData[i], j);
wts[j] += delWT * LearningRate;
});
if (OnTraining != null)
{
TrainingResponse res = new TrainingResponse();
res.Loss = 0.5f * System.Math.Pow((dynamic)knownOutputs[i] - Predict(inputData[i]), 2);
OnTraining?.Invoke(this, res);
}
});
Parallel.For(0, Weights.Count, (j) =>
{
Weights[j] += wts[j] * LearningRate;
});
//Slope = Slope + delta_m * LearningRate;
//Bias = Bias + delta_c * LearningRate;
}
void SGD(T[] inputData, T[] knownOutputs)
{
Parallel.For(0, inputData.Length, (i) =>
{
double delta_m = inputData[i] * ((dynamic)knownOutputs[i] - Predict(inputData[i])) * DSigmoid((inputData[i]));
double delta_c = ((dynamic)knownOutputs[i] - Predict(inputData[i]));
Slope += delta_m * LearningRate;
Bias += delta_c * LearningRate;
if (OnTraining != null)
{
TrainingResponse res = new TrainingResponse();
res.Loss = 0.5f * System.Math.Pow((dynamic)knownOutputs[i] - Predict(inputData[i]), 2);
OnTraining?.Invoke(this, res);
}
});
}
void SGD(T[] inputData, T[] knownOutputs)
{
Parallel.For(0, inputData.Length, (i) =>
{
Parallel.For(0, Weights.Count, (j) =>
{
double delWT = ((dynamic)knownOutputs[i] - Predict(inputData[i])) * System.Math.Pow((double)(dynamic)inputData[i], j);
Weights[j] += delWT * LearningRate;
});
if (OnTraining != null)
{
TrainingResponse res = new TrainingResponse();
res.Loss = 0.5f * System.Math.Pow((dynamic)knownOutputs[i] - Predict(inputData[i]), 2);
OnTraining?.Invoke(this, res);
}
});
}