The goal of this project is to provide a set of helper classes and functions for Theano to facilitate experiments with machine learning, in particular for natural language processing. Unlike libraries such as Keras and Blocks, the purpose is rather to help people (myself in particular) use Theano efficiently when developing new architectures.

BNAS is used by and developed in tandem with HNMT, the Helsinki Neural Machine Translation (NMT) system. A simpler NMT system for educational purposes can be found in the examples/ directory.

Since this is a project currently under development, you probably want to install it in development mode:

python3 setup.py develop --user

The examples/ directory contains commented example code that you could start from.

Install Sphinx and run docs/build.sh to build the API reference.