package torch

create vs ~input_dim ~hidden_size creates a new RNN with the specified input dimension and hidden size.

step t state input_ applies one step of the RNN on the given input using the specified state. The updated state is returned.

seq t inputs ~is_training applies multiple steps of the RNN starting from a zero state. The hidden states and the final state are returned. inputs should have shape batch_size * timesteps * input_dim, the returned output tensor then has shape batch_size * timesteps * hidden_size.

zero_state t ~batch_size returns an initial state to be used for a RNN.