Uncertainty-Aware Text-to-Program for Question Answering on Structured Electronic Health Records (CHIL 2022)

by Daeyoung Kim (KAIST), Seongsu Bae (KAIST), Seungho Kim (KAIST), Edward Choi (KAIST)

This repository provides the official implementation of the Uncertainty-Aware Text-to-Program for Question Answering on Structured Electronic Health Records.

Requirements

• PyTorch == 1.7.1
• Python == 3.8.5
• transformers == 4.5.1
• numpy == 1.19.5
• pytorch-lightning == 1.3.2
• rdflib == 5.0.0

Data

Prepare Knowledge Graph

You should build knowledge graph for MIMICSPARQL* following instruction in official MIMICSPARQL* github.
The KG(mimic_sparqlstar_kg.xml) file should be in ./data/db/mimicstar_kg directory.

Pre-process

Generate dictionary files for the recovery technique.

$ cd data
$ python preprocess.py

Train

$ python main.py

Test

$ python main.py --test

Citation

@article{kim2022uncertainty,
  title={Uncertainty-Aware Text-to-Program for Question Answering on Structured Electronic Health Records},
  author={Kim, Daeyoung and Bae, Seongsu and Kim, Seungho and Choi, Edward},
  journal={arXiv preprint arXiv:2203.06918},
  year={2022}
}