This repository replicates the results from Zarembra et al. (2014). There are a couple of minor differences such as replacing their custom initialization scheme which was removed since it makes no discernible improvement over TensorFlow's default Xavier initialization.
It is based off of the official TensorFlow tutorial for RNNs but contains approximately half the lines of code.
- TensorFlow 1.2 (may work with 1.1 but this has not been tested)
- NumPy