SEASS

This repository contains code for the ACL 2017 paper "Selective Encoding for Abstractive Sentence Summarization"

About this code

The experiments in the paper were done with an in-house deep learning tool. Therefore, we re-implement this as a reference.

PyTorch version: This code requires PyTorch v0.3.x.

Python version: This code requires Python3.

How to run

Prepare the dataset and code

You can download the processed data from here. Or, you can process the dataset with NAMAS.

Make a folder for the code and data:

SEASS_HOME=~/workspace/seass
mkdir -p $SEASS_HOME/code
cd $SEASS_HOME/code
git clone --recursive https://github.com/magic282/SEASS.git

Put the data in the folder $SEASS_HOME/code/data/giga and organize them as:

seass
├── code
│   └── SEASS
│       └── seq2seq_pt
└── data
    └── giga
        ├── dev
        ├── models
        └── train

Since the validation set is large, you can sample a small set from it.

Collect vocabulary using CollectVocab.py. Then put the vocab files in the train folder.

Modify run.sh according to your setting and files.

Setup the environment

Package Requirements:

nltk scipy numpy pytorch

Warning: Older versions of NLTK have a bug in the PorterStemmer. Therefore, a fresh installation or update of NLTK is recommended.

A Docker image is also provided.

Docker image

docker pull magic282/pytorch:0.3.1

Run training

The file run.sh is an example. Modify it according to your configuration.

Without Docker

bash $SEASS_HOME/code/SEASS/seq2seq_pt/run.sh $SEASS_HOME/data/giga $SEASS_HOME/code/SEASS/seq2seq_pt

With Docker

nvidia-docker run --rm -ti -v $SEASS_HOME:/workspace magic282/pytorch:0.3.1

Then inside the docker:

bash code/SEASS/seq2seq_pt/run.sh /workspace/data/giga /workspace/code/SEASS/seq2seq_pt