Invariant Rationalization

This repo contains the Tensorflow implementation of Invariant Rationalization on the polluted IMDB dataset (please refer to the paper for details) for the demo purpose.

The original beer review dataset has been removed by the dataset’s original author, at the request of the data owner, BeerAdvocate. To respect the wishes and legal rights of the data owner, we do not include the data in our repo. If you are interested in beer review, please first obtain the dataset from the original authors who released the dataset. We will then be happy to provide our data and environment partitions to whoever is granted rights to the data. Once you have access to the dataset and have prepared it in the desired format, please refer to the beer_review branch of this repo.

There is a short video that explains the main concepts of our work. If you find this work useful and use it in your research, please consider citing our paper.

@article{chang2020invariant,
  title={Invariant Rationalization},
  author={Chang, Shiyu and Zhang, Yang and Yu, Mo and Jaakkola, Tommi S},
  journal={arXiv preprint arXiv:2003.09772},
  year={2020}
}

Getting Started

Below is the step-by-step instruction on running our code. After cloning the repo, please first run the following script to download the dataset and the pre-trained word embeddings.

sh download_data.sh

The IMDB dataset downloaded here is identical to the one in tensorflow datasets. You could of course directly using the build-in functions in TensorFlow with some minor tweaks.

Next, you need to properly set up your virtual environment and install all dependencies from the requirements.txt. You may consider the following commend:

pip install -r requirements.txt

Once all packages are installed, now you are ready to have some fun by using the following commends under your main directory.

sh run_demo.sh

You are expected to see a similar result as the following.

The best dev performance appears at epoch: 28.
The best dev performance and bias selection are: 0.8454, and 0.0000.
The corresponding test performance and bias selection are: 0.8492, and 0.0000.

Tested environment: Python 3.6.10, Tensorflow: 1.14, CUDA driver: 410.48, CUDA version: 10.0, and Cudnn: 7.4.1.

Final Words

That's all for now and hope this repo is useful to your research. For any questions, please create an issue and we will get back to you as soon as possible.