Learning Recurrent Span Representations for Extractive Question Answering

https://arxiv.org/abs/1611.01436

Requirements

Theano, Matplotlib, Java

Initial setup

$ python setup.py

This will download GloVe word embeddings and tokenize raw training / development data.
(download will be skipped if zipped GloVe file is manually placed in data directory).

Training

$ python rasor.py --device DEVICE --train

where DEVICE is cpu, or an indexed GPU specification e.g. gpu0.
When specifying a certain GPU, the theano device flag must be set to cpu, i.e. set device=cpu in your .theanorc file.

Making predictions

$ python rasor.py --device DEVICE test_json_path pred_json_path

where test_json_path is the path of a JSON file containing articles, paragraphs and questions (see SQuAD website for specification of JSON structure), and pred_json_path is the path to write predictions to.

---

Tested in the following environment:

• Ubuntu 14.04
• Python 2.7.6
• NVIDIA CUDA 8.0.44 and cuDNN 5.1.5
• Theano 0.8.2
• Matplotlib 1.3.1
• Oracle JDK 8