data-driven-missingness-cru-irregular-timeseries

Code for irregular time-series models with data-driven missingness assumption

Cloning the anonymous repo

Please follow the instructions here to clone an anonymous repo. (Credit : Clone Anonymous Github created by fedebotu)

Run toy experiments

Run the following lines of code for reproducing the toy experiments

Create the environment

conda create -n data-driven-miss-cru python=3.9.7

conda activate data-driven-miss-cru

pip install -r requirements.txt

CRU-FM

python CRU/run_toy_experiment_cru.py --random_seed 68 --mnar True

CRU

python CRU/run_toy_experiment_cru.py --random_seed 68 --mnar False

• The results will be saved in the "CRU/training_results/toy_mnar/test_true_vs_predicted*mnar=True/False*.png"

Please let the CRU model run for atleast 100 epochs (default). It should take no more than 8-10 minutes on any machine.

mTAND

bash mTAND/run_toy_mnar_extrapolation.sh

• The results will be saved in mTAND/results folder

LatentODE

bash LatentODE/run_toy_mnar_experiment.sh

• The results will be saved in LatentODE/results folder

NeuralCDE

bash NeuralCDE/run_toy_mnar_experiment.sh

• The results will be saved in NeuralCDE/results folder

pVAE

bash pVAE/run_toy_mnar_extrapolation.sh

• The results will be saved in NeuralCDE/results folder

Note : To run the pVAE experiment, please create a separate enviroment using this requirements.txt file.

Run MIMIC-IV experiments

-Follow the instructions to pre-process the data in data_preprocessing/MIMIC-IV

Then run the script to train CRU-FM and CRU bash CRU/launch_cru_extrapolation_mimic.sh run_here

Run eICU experiments

-Follow the instructions to pre-process the data in data_preprocessing/eICU

Then run the script to train CRU-FM and CRU bash CRU/launch_cru_extrapolation_eicu.sh run_here