brchung2 / TPC-LoS-prediction

Repository for DLH Replication Project:

Citation to the original paper: Rocheteau, E., Lio`, P. & Hyland, S. Temporal point- wise convolutional networks for length of stay pre- diction in the intensive care unit. In Proceedings of the Conference on Health, Inference, and Learning, 58–68 (2021). https://arxiv.org/abs/2007.09483

Link to the original paper’s repo (if applicable): https://github.com/EmmaRocheteau/TPC-LoS-prediction

Dependencies: Package Dependencies in: requirements.txt

Data download instruction: Need credentialed access to eICU dataset.

1. To run the sql files, set up eICU database: https://physionet.org/content/eicu-crd/2.0/.

2. Follow the instructions: https://eicu-crd.mit.edu/tutorials/install_eicu_locally/ to ensure the correct connection configuration.

Preprocessing code + command (if applicable):

Preprocessing code in : eICU_preprocessing/create_all_tables.sql.

Commands:

1. Clone this repository

2. Replace the eICU_path in paths.json to a convenient location in your computer, and do the same for eICU_preprocessing/create_all_tables.sql using find and replace for '/content/drive/MyDrive/eICU_data/'. Leave the extra '/' at the end.

3. In your terminal, navigate to the project directory, then type the following commands:

   psql 'dbname=eicu user=eicu options=--search_path=eicu'
   

   Inside the psql console:

   \i eICU_preprocessing/create_all_tables.sql
   

   This step might take a couple of hours.

   To quit the psql console:

   \q
   

4. Then run the pre-processing scripts in your terminal. This may take a couple hours:

   python3 -m eICU_preprocessing.run_all_preprocessing
   

Training and Evaluation code + command (if applicable):

Training and Evaluation code for each model listed separately: eg models/run_tpc.py to run the TPC model; models/run_lstm.py to run the LSTM model.

1. Set the working directory to the TPC-LoS-prediction, and run the following command:

Specify command line arguments to hyperparameter values:

```

python3 -m models.run_tpc --dataset eICU --task LoS --model_type tpc --percentage_data 10  #TPC model
python3 -m models.run_LSTM --dataset eICU --task LoS --model_type lstm --percentage_data 10 #LSTM model
python3 -m models.run_transformer --dataset eICU --task LoS --model_type transformer --percentage_data 10 #Transformer model  

```

2. Run this script to run experiments:

   python3 -m models.hyperparameter_scripts.eICU.hyperparameter_tuning_exp 
   

Pretrained model (if applicable):

Table of results (no need to include additional experiments, but main reproducibility result should be included)

Presentation slides: https://docs.google.com/presentation/d/1-GQfAKFpgfdcnkTdKmeyAoaGeWNCnBUxEgQbWf3nc6I/edit#slide=id.g222d1ba3356_0_134

Performance for each model on 10% data and the improvement range % of the TPC model over the best baseline. Bolded metric numbers show the best model is TPC for each metric.