GitHub - Billijk/DeepQuestionAsking (original) (raw)
Modeling Question Asking Using Neural Program Generation
Code for our CogSci 2021 paper Modeling Question Asking Using Neural Program Generation (ArXiv).
Requirements:
- Python3
- PyTorch >= 1.1.0
- expected-information-gain
How to run our code
Estimating the distribution of human questions
This needs to pre-train the model on the synthesized dataset and the fine-tune on the human questions.
For cross-validation, run
./run_human.sh validation
Question generation
For inference and evaluation, run
./run_gen.sh eval ./checkpoints/ep_500.pth
Acknowledgement
This work was supported by Huawei. We are grateful to Todd Gureckis and Anselm Rothe for helpful comments and conversations. We thank Jimin Tan for writing the initial version of the RL-based training codes.
Citation
If you use our codes, please kindly cite our paper.
@inproceedings{wang2021modeling, title={Modeling Question Asking Using Neural Program Generation}, author={Wang, Ziyun and Lake, Brenden}, booktitle={Proceedings of CogSci 2021}, year={2021} }