Hi Emebet,
A Nonlinear Autoregressive with Exogenous Input (NARX) model uses past values of both the input signals and its own outputs to make predictions about future values.
- The notation ''net.inputWeights{i,j}" refers to the weight matrix that connects the inputs to the neurons in layer i. The {i,j} notation specifies the connection from the j-th input to the i-th layer. For a NARX model with two time steps, you can have "net.inputWeights{1,1}" for the weight associated with the first delay and ''net.inputWeights{1,2}" for the weight related to the second delay. The statement ''net.inputWeights{i,j} = 2" is not a standard assignment; rather, these weights are learned during the model training process.
- The property ''net.inputWeights{i,j}.delays = [1,2]" specifies which past time steps of the input signals are being utilized in the model. In this case, it indicates that the model incorporates values from one time step ago and two time steps ago. This allows the network to leverage historical data effectively, enhancing its predictive capabilities.
I hope this clears the query.
For more details refer to the MathWorks documentation on 'NARX': https://www.mathworks.com/help/deeplearning/ref/narxnet.html
