The dataset used is the MIMIC III dataset, fount at https://mimic.physionet.org.
STEP I: install dependencies
pip install -r requirements.txt
STEP II: data extraction & preprocessing
python scr/data_processing/main.py [-h] -u SQLUSER -pw SQLPASS -ht HOST -db DBNAME -r SCHEMA_READ_NAME [-w SCHEMA_WRITE_NAME]
STEP III: run the model
├── LICENSE
├── README.md <- The top-level README for developers using this project.
├── data
│ ├── external <- Data from third party sources.
│ ├── interim <- Intermediate data that has been transformed.
│ ├── processed <- The final, canonical data sets for modeling.
│ ├── raw <- The original, immutable data dump.
│ ├── train <- The training data used for ... training.
│ ├── val <- The validation data used for ... validating (and hyperparameter selection).
│ └── test <- The test data used for reporting.
│
├── src <- Source code for use in this project.
│ ├── __init__.py <- Makes src a Python module
│ │
│ ├── mains <- Runs the full pipeline
│ │ └── GP_TCN_stat_main.py <- in use for MGP-TCN; MGP-AttTCN
│ │
│ ├── data_loader <- Loads the data into main
│ │ └── raw_irreg_loader.py <- in use for MGP-TCN; MGP-AttTCN
│ │
│ ├── models <- Models to load into main
│ │ ├── GP_TCN_Moor.py <- re-implementation of Moor et. al. (MGP-TCN)
│ │ └── GP_attTCN.py <- thesis model: MGP + attention based TCN (MGP-AttTCN)
│ │
│ ├── trainer <- Trains the data
│ │ └── GP_trainer_with_stat.py <- in use for MGP-TCN; MGP-AttTCN
│ │
│ ├── loss_n_eval <- Files to calculate loss, gradients and AUROC, AUPR
│ │ └── ...
│ │
│ ├── visualization <- Scripts to create exploratory and results oriented visualizations
│ │
│ ├── data_preprocessing <- Scripts to download or generate data
│ │
│ └── features_preprocessing <- Scripts to turn raw data into features for modeling
│
└── requirements.txt <- The requirements file for reproducing the analysis environment, e.g.
generated with `pip freeze > requirements.txt`
Credits to M. Moor for sharing his code from https://arxiv.org/abs/1902.01659