PubmedQA system

This repo contains the code for BioQANet: Enhancing Biomedical Research Question Answering with Transformer Models

Team Members:

Repo structure

data : contains the data required for experiments. Create the folder and place data from original PubMedQA here
preprocess : contains the original script from PubmedQA for splitting expert into train and test
predictions : contains the predicted json files
Archived notebooks : Self explanatory

EDA.ipynb : Basic exploratory analysis and simple baselines for QA model
Contrastive-512.ipynb : Notebook for training and inference of BioQANet
Contrastive-without-pretraining.ipynb : Notebook for training and inference without pretraining
Contrastive.ipynb : Similar to Contrastive-512 but has max length 400

utils.py : Utility script
contrastive-utils.py: Functions for model definition used in Contrastive-512 notebook
evaluation.py: Original function from PubmedQA for evaluating the predictions
get_human_performance.py: Original function from PubmedQA for getting human performance metrics

python evaluation.py path_to_pred.json

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
Archived notebooks		Archived notebooks
predictions		predictions
preprocess		preprocess
EDA.ipynb		EDA.ipynb
README.md		README.md
config.py		config.py
contrastive-512.ipynb		contrastive-512.ipynb
contrastive-without-pretraining.ipynb		contrastive-without-pretraining.ipynb
contrastive.ipynb		contrastive.ipynb
contrastive_utils.py		contrastive_utils.py
evaluation.py		evaluation.py
get_human_performance.py		get_human_performance.py
utils.py		utils.py