<!-- ![alt-text-1]("title-1")  -->

The implementation of NAACL 2021 paper "Event Representation with Sequential, Semi-Supervised Discrete Variables".

<img src="figs/model.png" width="600" height="200"/><img src="figs/example.png" width="400"/>

This Pytorch code implements the model and reproduces the results from the paper.

Conda Environment:

conda create --name ssdvae
conda install pytorch==0.4.1 cuda92 -c pytorch
pip install torchtext==0.2.3
conda install pandas scikit-learn
conda install -c conda-forge pytablewriter

Data:

Wikipedia Dataset

Wikipedia Inverse Narrative Cloze

NYT Inverse Narrative Cloze

Training:

./train.sh $obsv_prob $exp_num $seed

data_mode:

{'valid','test'}

Perplexity:

./test_ppx.sh $obsv_prob $exp_num $seed $data_mode

Wiki Inverse Narrative Cloze:

./wiki_inv_narr.sh $obsv_prob $exp_num $seed $data_mode

NYT Inverse Narrative Cloze:

./nyt_inv_narr.sh $obsv_prob $exp_num $seed $data_mode

Disclaimer

Some parts of the code were inspired by HAQAE implementations.