public void Train(double[] input, double[] labelArray)
{
// Making prediction
var inputs = Matrix <double> .Build.Dense(input.Length, 1, (i, j) => input[i]);
var hidden = _weightsInHidden.Multiply(inputs);
hidden.Add(_biasHidden, hidden);
hidden.MapInplace(val => _activation.Activasion(val));
var output = _weightsHiddenOut.Multiply(hidden);
output.Add(_biasOut, output);
output.MapInplace(val => _activation.Activasion(val));
// Building target matrix
var target = Matrix <double> .Build.Dense(labelArray.Length, 1, (i, j) => labelArray[i]);
// Calculate error
// Error => Target - output
var errors = target.Subtract(output);
// Gradient = outputs * (1 - outputs);
// Calculate gradient
var gradiants = output.Map(val => _activation.DActivasion(val));
InPlaceMultiply(ref gradiants, ref errors);
gradiants.Multiply(_learningRate, gradiants);
// Calculate deltas
var hidden_t = hidden.Transpose();
var weight_hidden_out_deltas = gradiants.Multiply(hidden_t);
//lock (_wHiddenOutLock)
//{
// Adjust weights by deltas
_weightsHiddenOut.Add(weight_hidden_out_deltas, _weightsHiddenOut);
// Adjust bias by deltas, which is gradiant
_biasOut.Add(gradiants, _biasOut);
//}
// Calculate hidden layer errors
var weights_hidden_out_transposed = _weightsHiddenOut.Transpose();
var hidden_errors = weights_hidden_out_transposed.Multiply(errors);
// Calculate hidden gradiant
var hidden_gradiant = hidden.Map(val => _activation.DActivasion(val));
InPlaceMultiply(ref hidden_gradiant, ref hidden_errors);
hidden_gradiant.Multiply(_learningRate, hidden_gradiant);
// Calculate input => hidden deltas
var inputs_t = inputs.Transpose();
var weights_input_hidden_deltas = hidden_gradiant.Multiply(inputs_t);
//lock (_wInHiddenLock)
//{
_weightsInHidden.Add(weights_input_hidden_deltas, _weightsInHidden);
_biasHidden.Add(hidden_gradiant, _biasHidden);
//}
}
