This repo contains the code for the Structured model for CoNaLa dataset, as described in the paper: https://aclanthology.org/2021.findings-acl.384.pdf
- Download the data from http://conala-corpus.github.io
- Build a file with NL intents of training set: Refer to datasets/conala/retrive_src.py. The file will be saved as src.txt in data/conala
- Parse those NL intents and build a vocabulary: refer to https://github.com/nxphi47/tree_transformer for more details on setting up the parser and run convert_ln.py
- Build train/dev/test dataset: Run datasets/conala/dataset_hie.py
- Train the model: scripts/conala/train.sh
- Test the model: scripts/conala/test.sh
This repo is built on top of two projects: a. TRANX: https://github.com/pcyin/tranX b. tree_transformer: https://github.com/nxphi47/tree_transformer
So please cite their work.
@inproceedings{dahal-etal-2021-analysis,
title = "Analysis of Tree-Structured Architectures for Code Generation",
author = "Dahal, Samip and
Maharana, Adyasha and
Bansal, Mohit",
booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
month = aug,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.findings-acl.384",
doi = "10.18653/v1/2021.findings-acl.384",
pages = "4382--4391",
}