ProCare

This repository contains a demo implementation of our ProCare.

Please note that if the local pytorch contains the cpu version, it may cause an error to be reported, which is a library issue

Environment Setup

1. Pytorch 1.12.1

2. Python 3.7.15

3. torch-scatter: pip install torch-scatter -f https://pytorch-geometric.com/whl/torch-1.12.1+cu102.html

4. torch-sparse: pip install torch-sparse -f https://pytorch-geometric.com/whl/torch-1.12.1+cu102.html

5. torch-geometric: pip install torch-geometric

6. torchdiffeq: pip install torchdiffeq

Guideline

data

We provide one implementation on MIMIC-III dataset.

binary_train_codes_x.pkl train set

binary_test_codes_x.pkl test set

train_codes_y.npy train label

test_codes_y.npy test label

train_visit_lens.npy the number of visits per patient in train set

test_visit_lens.npy the number of visits per patient in test set

code_levels.npy ICD-9 disease code tree structure

patient_time_duration_encoded.pkl timestamp of all patient visits

sum_TE2.0.pkl the severity-driven time embedding

train_pids.npy the mapping of patient's pids to idx in train set

test_pids.npy the mapping of patient's pids to idx in test set

models

The implementation of model(ProCare.py);

Example to run the codes

python ProCare.py