andreev-io/PACE

End-to-end replication of the Machine Learning paper on task decomposition in healthcare

Replicating PACE - Learning Effective Task Decomposition for Human-in-the-loop Healthcare Delivery

Citation

Kaiping Zheng, Gang Chen, Melanie Herschel, Kee Yuan Ngiam, Beng Chin Ooi, and Jinyang Gao. 2021. PACE: Learning Effective Task Decomposition for Human-in-the-loop Healthcare Delivery. Proceedings of the 2021 International Conference on Management of Data. Association for Computing Machinery, New York, NY, USA, 2156–2168. https://doi.org/10.1145/3448016.3457281

Replication Instructions

1. Set up a conda environment with the packages defined in the provided environment.yml, and activate the environment

conda env create --name pace-replication --file=environment.yaml
conda activate pace-replication

2. Download the preprocessed dataset built using MIMIC Extract from GCP: https://console.cloud.google.com/storage/browser/mimic_extract. To access this, you will need access to the MIMIC-III dataset, and link your physionet account to GCP.

3. Run the code in mimic_extract_analysis.ipynb, to process the downloaded file, and generate training, validation and test datasets.

4. Run the code in model.ipynb to train the model, run tests, and save the predictions and trained model to a local directory.

5. Run the code in metrics.ipynb to obtain the graphs used in the report

Dependencies

Listed in environment.yml

Table of results

The predictions used in the paper are included in the predictions directory, which is used in metrics.ipynb. The reference labels for these labels are provided in test_labels.pkl.