keraflow.layers.recurrent.Recurrent Class Reference

Base class for recurrent layers. More...

Inheritance diagram for keraflow.layers.recurrent.Recurrent:

Public Member Functions
def	__init__

Detailed Description

Base class for recurrent layers.

Do not use this layer in your code.

Constructor & Destructor Documentation

def keraflow.layers.recurrent.Recurrent.__init__	(		self,
			num_states,
			output_dim,
			return_sequences = False,
			go_backwards = False,
			stateful = False,
			unroll = False,
			kwargs
	)

Parameters

num_states	int. Number of state of the layer.
output_dim	int. The output dimension of the layer.
return_sequences	Boolean. Whether to return the last output in the output sequence, or the full sequence.
go_backwards	Boolean. If True, process the input sequence backwards.
stateful	Boolean. See below.
unroll	Boolean. If True, the network will be unrolled, else a symbolic loop will be used. When using TensorFlow, the network is always unrolled, so this argument has no effect. Unrolling can speed-up a RNN, although it tends to be more memory-intensive. Unrolling is only suitable for short sequences.
kwargs	see Layer.__init__.

Note

1. You can set RNN layers to be stateful by setting stateful=True, which means that the states computed for the samples in one batch will be reused as initial states for the samples in the next batch. This assumes a fixed batch size (specified by Input) and a one-to-one mapping between samples in different successive batches.
2. All recurrent layer supports masking for input data with a variable number of timesteps. To introduce masks to your data, specify mask_value of Input. Usually you will concatenate an Input layer, an embedding layer, and then a recurrent layer.