Hybrid-ODE-NeurIPS-2021

Code for Integrating Expert ODEs into Neural ODEs: Pharmacology and Disease Progression (NeurIPS 2021).

Installation

Python 3.6+ is recommended. Install dependencies as per requirements.txt.

If CUDA support is needed:

• make sure you have appropriate drivers installed,
• make sure you have CUDA toolkit (a version compatible with PyTorch 1.10, see here or here) installed on your system or in you virtual environment.

Replicating Experiments

Shell scripts to replicate the experiments can be found in experiments/.

To run all the synthetic data experiments:

$ bash experiments/run_all.sh

You may also run the experiment steps individually, see experiments/run_all.sh. To then produce the figures, run the Jupyter notebooks Fig3.ipynb, Fig6.ipynb, Fig7.ipynb, Fig9.ipynb found under experiments/.

To run real data experiments:

1. Access to Dutch Data Warehouse dataset is required, see real_data/README.md for more information.

2. Preprocess the data, as documented in real_data/README.md.

3. Run experiments:

   $ bash experiments/real.sh

Citing

If you use this code, please cite the associated paper:

@inproceedings{NEURIPS2021,
  author = {Qian, Zhaozhi and Zame, William R and Fleuren, Lucas M and Elbers, Paul and van der Schaar, Mihaela},
  booktitle = {Advances in Neural Information Processing Systems},
  title = {Integrating Expert ODEs into Neural ODEs: Pharmacology and Disease Progression},
  url = {https://papers.neurips.cc/paper/2021/file/5ea1649a31336092c05438df996a3e59-Paper.pdf},
  volume = {34},
  year = {2021}
}