This is the code for the TKDE submission paper: Stage-Aware Hierarchical Attentive Relational Network for Diagnosis Prediction. We have implemented our methods in Pytorch and DGL. Here, we provide the source code for model-variant HAR-LSTM.
You need to download MIMIC-III and MIMIC-IV datasets by yourself. Then you need finish data-processing as following:
cd data_util;
python simplify_semmed.py
python prepare_icd2cui.py
python filter_semmed_kg_edges.py
python process_mimic.py
Then you can run the file main.py
to train the model.
## Requirements
- Python 3
- PyTorch
- DGL