nn.RNN(input_size, hidden_size, num_layers=1, nonlinearity=tanh, bias=True, BATch_first=False, dropout=0, bidirectional=False)