So I have a couple questions about the design of a neural network. I'm trying to create a neural network to predict the number of fantasy points a player will score in a given week.
First of all, I started with a regular NN and the results were okay but not great. I'd like to try with an RNN as I see this as a time series problem, which RNNs are typically used for. I'd like to take a players last n games (probably around a season's worth) and predict their next game.
However when I switch players in my input and start predicting on a new player, I'm worried that the previous players will be remembered by the RNN when predicting the new player. Essentially each player is a new time series. Is there a proper way to train the neural network so that this isn't an issue? Would it even be an issue? I've used RNNs before on text but the entire passage was continuous so I didn't have this problem.
Also, I have a lot of data that isn't part of the time series, such as the upcoming opposing team. How should I incorporate this into the neural network?