I am dealing with the nnet package in R.
I know that the momentum $\alpha$ is used to decrease the fluctuations in weight changes over consecutive iterations. The original update of the weights states that:
$$\omega_i(t+1) = \omega_i - \eta\frac{\partial E}{\partial w_i},$$ where $E({\bf w})$ is the error function, ${\bf w}$ - the vector of weights and $\eta$ - learning rate.
The weight decay $\lambda$ penalizes the weight changes in the following way:
$$\omega_i(t+1) = \omega_i - \eta\frac{\partial E}{\partial w_i} - \lambda\eta\omega_i$$
Now, what I see on the nnet package is that I can control the $\lambda$ through the parameter "decay". How can I control the learning rate ($\eta$) instead?
