rnn : Recurrent Library for Torch

TL;DR

The rnn package for Torch offers flexible, tested components for building various Recurrent Neural Networks, enhancing capabilities while maintaining compatibility and supporting research comparisons.

Contribution

It introduces a comprehensive, flexible RNN library within Torch, with strong testing and backward compatibility, improving upon previous implementations.

Findings

Compared favorably with existing RNN implementations

Supports multiple RNN architectures and iterations

Ensures robustness through extensive testing

Abstract

The rnn package provides components for implementing a wide range of Recurrent Neural Networks. It is built withing the framework of the Torch distribution for use with the nn package. The components have evolved from 3 iterations, each adding to the flexibility and capability of the package. All component modules inherit either the AbstractRecurrent or AbstractSequencer classes. Strong unit testing, continued backwards compatibility and access to supporting material are the principles followed during its development. The package is compared against existing implementations of two published papers.