netket.experimental.nn.rnn.RNNCell

class netket.experimental.nn.rnn.RNNCell

Bases: Module

Recurrent neural network cell that updates the hidden memory at each site.

Attributes

features: int

output feature density, should be the last dimension.

Methods

abstractmethod __call__(inputs, cell_mem, hidden)

Applies the RNN cell to a batch of input sites at a given index.

Parameters:

• inputs – input data with dimensions (batch, in_features).

• cell_mem – cell memory from the previous site with dimensions (batch, features).

• hidden – hidden memories from the previous neighbors with dimensions (batch, n_neighbors, features).

Returns:

• cell_mem

  the updated cell memory with dimensions (batch, self.features).

• outputs

  the updated hidden memory with dimensions (batch, self.features), also serves as the output data at the current site for the netket.experimental.nn.rnn.RNNLayer layer.