TensorFlow version number: 1.1.0
tf.contrib.rnn.BasicRNNCell ()
__init__(
num_units,
input_size=None,
state_is_tuple=True
)
num_numits refers to how many hidden layer units there are;
activation defaults to the tanh activation function.
__call__(
inputs,
state,
scope=None
)
In the call function, inputs refer to the hidden layer input, and state refers to the hidden layer state of the previous time, so
output = new_state = activation(W * input + U * state + B)
About BasicRNNCell, there is a detailed introduction in the blog https://saicoco.github.io/tensorflow2/ .