In this file I added two layers ( keras - tensorflow ) that can be used in the following manner: layer_1 = half_and_half_layer(layer_0, args**)
Note that this is not like the typical functional layers in which you end the line with "(layer_0)"
The two layer implemented are :
- a convolutional highway layer
- a convolutional glu layer that is contrained to passing exactly half of the information onwards.
In practice the half_and_half_layer ( glu layer ) is as effective as highway layer and works much much faster.
Cheers, Dan