Procedure FFNN: |
Input Data: · x(t) is a time series of k observations · y(t) input-output mapping value |
Define Hyper-Parameters: · Epoch · Learning rate (η) · Batch Size · Class weight · Callbacks · Dense layer · Dropout layer |
Initialize Weight · wi: Initial input weight to random value |
Define the Cost Function: · Same as in Equation (9) |
For i in Range(Epoch): · Calculate E for all inputs. If E is smaller than tolerated value then exit from the loop with raising an exception · For each input, calculate gradients for all weights, included bias weights · If length of gradient vector is smaller than given minimum border value, then raise an exception and exit from the loop. · Else modify all weights by adding a negative multiple of the gradient to the weights, calculate accuracy and loss for the given epoch. End if End loop |
End Procedure |