Why did you choose a specific neural network architecture for this time-series data instead of an alternative approach, and what were the tradeoffs?
