This repository contains code needed to reproduce results from the paper of the same name.

This project uses Python 3.6.9. Notable dependencies are:

• tensorflow 1.10
• numpy 1.14.5
• matplotlib
• scikit-learn

A complete list of packages may be found in requirements.txt, and may be installed as follows:

pip install -r requirements.txt

Models and training scripts may be found in the src folder. If you have questions, please open an issue or email m.engelhard@duke.edu

Results on the synthetic dataset we describe may be obtained by running src/model.py. Results for baselines models may be obtained by running src/baselines.py.

Reddit data must first be downloaded through the pushshift.io API. Results may then be obtained by running src/train_reddit.py; data formatting may be inferred from the load_batch function. Results for baseline models may be obtained by running src/reddit_baselines.py.

MIMIC data must first be requested and downloaded from https://mimic.physionet.org. Results may then be obtained by running src/train_mimic.py; data formatting may be inferred from the load_batch function. Results for baseline models may be obtained by running src/mimic_baselines.py.