RNNCell

source

RNNCell(
   input_layer, n_units, previous_state = None, activation = tf.tanh,
   w_init = tf.initializers.glorot_uniform(),
   u_init = tf.initializers.glorot_uniform(), bias_init = tf.initializers.zeros(),
   share_state_with = None, w_dropconnect = None, u_dropconnect = None,
   r_dropout = None, x_dropout = None, y_dropout = None, dropout_locked = True,
   regularized = False, name = 'rnn_cell'
)

---

Recurrent Cell Corresponds to a single step on an unrolled RNN network

Args

• input_layer : the input layer to the RNN Cell
• n_units : number of output units for this RNN Cell
• previous_state : a RNNCell from which we can extract output
• activation : activation function to be used in the cell
• share_state_with : a Layer with the same number of units than this Cell
• name : name for the RNN cell share_state_with (RNNCell or None):

Methods:

.init_state

source

.init_state()

.compute

source

.compute(
   input_layer, *previous_state
)