Neural Ordinary Differential Equations for Intervention Modeling (IMODE)

This repository is the official implemtation of Neural Ordinary Differential Equations for Intervention Modeling

Real-world systems often involves external interventions that cause changes in the system dynamics such as a patient being administered with particular drug. We propose a novel neural ODE-based approach (IMODE) that properly model the effect of external interventions by employing two ODE functions to separately handle the observations and the interventions.

A simulation of the trajectory in a 2D plane

Requirements

The below python3 packages are required to run the experiments.

torch
torchdiffeq
numpy
pandas
tqdm
matplotlib

Running the model

Exponential Decay

For training the model,

python main_decay.py --exp-name decay_exp

For testing the trained model,

python main_decay.py --exp-name decay_exp --test-phase True

Moving Ball

For training the model,

python main_collision.py --exp-name collision_exp 

For testing the trained model,

python main_collision.py --exp-name collision_exp --test-phase True

eICU

eICU Collaborative Research Database limits access by one who does not have a proper access, so the required training course must be completed prior to requesting access. Please see the following page to get the access https://eicu-crd.mit.edu/gettingstarted/access/.

For preprocessing eICU dataset, run eICU_preprocessor.ipynb in datasets/eICU folder.

For training the model,

python main_eicu.py --exp-name eicu_exp

For testing the trained model,

python main_eicu.py --exp-name eicu_exp --test-phase True