ASCC

Pytorch implementation of ICLR 2021 spotlight paper, "Towards Robustness Against Natural Language Word Substitutions"(https://openreview.net/forum?id=ks5nebunVn_). Our code is partly based on https://github.com/robinjia/certified-word-sub and https://github.com/JHL-HUST/PWWS .

Environment

conda env create --name ascc --file environment.yml

Spacy python -m spacy download en
GloVe cd glove wget http:https://nlp.stanford.edu/data/glove.840B.300d.zip unzip glove.840B.300d.zip rm glove.840B.300d.zip cd ..
IMDB cd data_set wget http:https://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz tar -xvzf aclImdb_v1.tar.gz rm -f aclImdb_v1.tar.gz cd ..
SNLI cd data_set wget https://nlp.stanford.edu/projects/snli/snli_1.0.zip unzip snli_1.0.zip rm -r snli_1.0.zip snli_1.0 __MACOSX cd ..

See train_imdb.sh

Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
PWWS		PWWS
from_certified		from_certified
lm_scores		lm_scores
models		models
solver		solver
LICENSE		LICENSE
README.md		README.md
counterfitted_neighbors.json		counterfitted_neighbors.json
dataHelper.py		dataHelper.py
environment.yml		environment.yml
label_smooth.py		label_smooth.py
opts.py		opts.py
requirements.txt		requirements.txt
train_imdb.sh		train_imdb.sh
train_imdb_ascc.py		train_imdb_ascc.py
utils.py		utils.py