logigan

LogiGAN

This repository serves primarily as codebase and data, model for training, evaluation and inference of the logical pre-training method LogiGAN. LogiGAN (NeurIPS 2022) is the adversarial logical pre-training method with Transformer-based encoder-decoder backbone.

The data and model are released in Here.

Preprocessing

Logic MLM Corpus Construction

cd corpus_construction/mlm_corpus
bash construct_premise.sh
bash construct_conclusion.sh

Elastic Search for External Negatives

cd corpus_construction/elastic_search
bash run_gen.sh
bash run_ver.sh

Adversarial Pretraining

Noting that the generator and verifier should be warmed up with constructed corpus to achieve better performance. Afterwards,

cd pre-training
#launcher the program, the setting of each step is adjusted in:
python launcher_es.py
(The parameters are adjusted in parameters16g_es_corpusb.py)

Citation

If you find this resource useful, please cite the paper introducing LogiGAN:

@article{pi2022logigan,
  title={LogiGAN: Learning Logical Reasoning via Adversarial Pre-training},
  author={Pi, Xinyu and Zhong, Wanjun and Gao, Yan and Duan, Nan and Lou, Jian-Guang},
  journal={arXiv preprint arXiv:2205.08794},
  year={2022}
}