sebastien-j

Currently, the `Bidirectional` class uses the same positional and keywords arguments for its two children. As such, non-recurrent weights and biases will generally be shared between the two networks, which...