Pass dropout once for whole neural-network

[aliciafmachado]

Currently, we have to set the dropout rate for each layer, and ideally we should only need to set it once.