README.md

Enhancing Tabular Reasoning with Pattern Exploiting Training

This repository contains the modified official code for the paper: "Enhancing Tabular Reasoning with Pattern Exploiting Training". Visit the project page at https://infoadapet.github.io/

ACL 2020 paper: INFOTABS: Inference on Tables as Semi-structured Data. To explore the dataset online visit project page.

Infotabs Dataset

The model improves and simplifies PET with a decoupled label objective and label-conditioned MLM objective.

How run on colab?

Open colab Make a new notebook and connect it to hardware accelerator.

!git clone https://github.com/abhilashreddys/infoADAPET.git
!pip install crcmod jsonpickle wandb transformers tqdm sentencepiece
cd infoADAPET/

Processed INFOTABS data is already in the git repo.
Download data (if required)
Superglue

mkdir -p data
cd data
mkdir -p superglue
cd superglue
wget "https://dl.fbaipublicfiles.com/glue/superglue/data/v2/combined.zip"
unzip combined.zip
cd ../..

Fewglue

mkdir -p data
cd data
git clone https://github.com/timoschick/fewglue.git
cd fewglue
rm -rf .git
rm README.md
mv FewGLUE/* .
rm -r FewGLUE
cd ../..

Set Env variables

%env PET_ELECTRA_ROOT= /content/infoADAPET
%env PYTHONPATH=$PET_ELECTRA_ROOT:$PYTHONPATH
%env PYTHON_EXEC=python

Training

!python -m src.train -c /content/ADAPET/config/{config_file}.json

Evaluation

!python -m src.dev -e /content/ADAPET/exp_out/fewglue/{task_name}/albert-xxlarge-v2/{timestamp}/
!python -m src.test -e /content/ADAPET/exp_out/fewglue/{task_name}/albert-xxlarge-v2/{timestamp}/

Contact

For any doubts or questions regarding the work, please contact Abhilash (sareddy53@gmail.com). For any bug or issues with the code, feel free to open a GitHub issue or pull request.