keraflow.layers.core.Highway Class Reference

Densely connected highway network. More...

Inheritance diagram for keraflow.layers.core.Highway:

Public Member Functions
def	__init__

Detailed Description

Densely connected highway network.

A natural extension of LSTMs to feedforward networks.

• input_shape: 2D, (nb_samples, input_dim)
• output_shape: 2D, (nb_samples, input_dim)
• parameters:
  • W, W_carry: (input_dim, input_dim)
  • b, b_carry: (input_dim,)
• Reference: Highway Networks

Constructor & Destructor Documentation

def keraflow.layers.core.Highway.__init__	(		self,
			transform_bias = -2,
			init = 'glorot_uniform',
			activation = 'linear',
			bias = True,
			kwargs
	)

Parameters

transform_bias	initial value for all elements of b_carry.
init	str/function. Function to initialize trainable parameters. See Initializations.
activation	str/function. Activation function applied on the output. See Activations.
bias	boolean. Whether to include a bias (i.e. make the layer affine rather than linear).
kwargs	see Layer.__init__.