• ManyDG: Many-domain Generalization for Healthcare Applications
• Data, Processing Scripts, Baselines and Model Codes

4-minute explanation video is provided in [YouTube] https://youtu.be/rbTBajnZV5U

• data/
  • drugrec/
    • this is the folder stores data for drug recommendation task
    • data_process.py: this file is used for processing MIMIC-III data for the task
  • Seizure/
  • sleep/
    • this is the data folder for the processed sample-based Sleep-EDF cassette portion.
    • sleep_edf_process.py: the processing file for sleep-edf (using multiprocess package for parallel programming)
    • for downloading the Sleep-EDF data, please refere to https://www.physionet.org/content/sleep-edfx/1.0.0/
  • hospitalization/
    • this is the folder for the processed eICU data
    • eICU_process_step1.py: the first processing step of eICU data
    • eICU_process_step2.py: the second processing step of eICU data
  • idxFile/
    • this folder stores the code mapping of different event types in eICU dataset. They will be generated from eICU_process_step1.py and eICU_process_step2.py. They are used to assign feature dimensions for hospitalization prediction task
• log/, pre-trained/
  • this two folders stores the automatically generated running logs and pre-trained models
• model.py
  • this file contains all the backbone and Base models for running the experiments
• run_drugrec/
  • model_drugrec.py: This file inherit the Base model and other model utility functions from model.py and support the drug recommendaton task
  • run_drugrec.py: This is the entry of drug recommendation task, specifiying the data loader and other initialization steps.
  • utils_drugrec.py: This file provides the data split and loading files.
• run_hospitalization/
  • model_hospitalization.py: This file inherit the Base model and other model utility functions from model.py and support the hospitalization prediction task
  • run_hospitalization.py: This is the entry of hospitalization prediction task, specifiying the data loader and other initialization steps.
  • utils_hospitalization.py: This file provides the data split and loading files.
• run_seizure/
  • model_seizure.py: This file inherit the Base model and other model utility functions from model.py and support the seizure detection task
  • run_seizure.py: This is the entry of seizure detection task, specifiying the data loader and other initialization steps.
  • utils_seizure.py: This file provides the data split and loading files.
• run_sleep/
  • model_sleep.py: This file inherit the Base model and other model utility functions from model.py and support the sleep staging task
  • run_sleep.py: This is the entry of sleep stagin task, specifiying the data loader and other initialization steps.
  • utils_sleep.py: This file provides the data split and loading files.

• For sleep staging

cd ./ManyDG
python data/sleep/sleep_edf_process.py
python run_sleep/run_sleep.py --model [MODEL] --cuda [WHICH GPU] --N_pat [N_OF_PAT] --epochs [EPOCHS]

usage: sleep_edf_process.py [-h] --model MODEL [--cuda CUDA] [--N_pat N_PAT] [--dataset DATASET] [--MLDG_threshold MLDG_THRESHOLD]
               [--epochs EPOCHS]

optional arguments:
  -h, --help            show this help message and exit
  --model MODEL         choose from base, dev, condadv, DANN, IRM, SagNet, PCL, MLDG
  --cuda CUDA           which cuda
  --N_pat N_PAT         number of patients
  --dataset DATASET     dataset name
  --MLDG_threshold MLDG_THRESHOLD
                        threshold for MLDG
  --epochs EPOCHS       N of epochs

• for seizure detection

cd ./ManyDG
# obtain the Seizure data first
python run_seizure/run_seizure.py --model [MODEL] --cuda [WHICH GPU] --N_vote [DEFAULT 5] --N_pat [N_OF_PAT] --epochs [EPOCHS]

• for drug recommendation

cd ./ManyDG
python data/drugrec/data_processing.py
python run_drugrec/run_drugrec.py --model [MODEL] --cuda [WHICH GPU] --N_pat [N_OF_PAT] --epochs [EPOCHS]

• for hospitalization prediction

cd ./ManyDG
python eICU_process_step1.py
python eICU_process_step2.py
python run_hospitalization/run_hospitalization.py --model [MODEL] --cuda [WHICH GPU] --N_pat [N_OF_PAT] --epochs [EPOCHS]