abductive-commonsense-reasoning

Public repository associated with Abductive Commonsense Reasoning, ICLR 2020

Setup

1. Clone the repo

   git clone https://github.com/allenai/abductive-commonsense-reasoning.git
   

2. Install requirements in a Python 3.7 environment

   pip install -r requirements.txt
   

3. Download Data and Models

   sh get-data.sh
   

The script will create data/ and models/ directories containing subdirectories anli and anlg for the Abuctive Inference and Generation tasks respectively. The size of the data and models is around 12G. If you'd like to download data:

wget https://storage.googleapis.com/ai2-mosaic/public/abductive-commonsense-reasoning-iclr2020/anli.zip
wget https://storage.googleapis.com/ai2-mosaic/public/abductive-commonsense-reasoning-iclr2020/anlg.zip

Tasks

1. Abductive Inference

   Interactive Demo

   python anli/demo.py --saved_model_dir models/bert-ft-lr1e-5-batch8-epoch4/ --gpu_id 0 --interactive
   

2. Abductive Generation

References

@inproceedings{
bhagavatula2020abductive,
title={Abductive Commonsense Reasoning},
author={Chandra Bhagavatula and Ronan Le Bras and Chaitanya Malaviya and Keisuke Sakaguchi and Ari Holtzman and Hannah Rashkin and Doug Downey and Wen-tau Yih and Yejin Choi},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=Byg1v1HKDB}
}