Learning Interpretable Hierarchical Dynamical Systems Models From Time Series Data [ICLR 2025]
We include the requirements.txt file to clone our python environment. Simply run
pip install -r requirements.txt
To start training a model, use the main.py file. Any hyper-parameters can be passed as command line arguments. Refer to the file, to see which hyper-parameters exist.
python main.py
Running multiple trainings, potentially in parallel, can be done via
python ubermain.py
any hyper-parameters must then be supplied in list format, see the file for more information. This also allows to do simple hyper-parameter grid searches.
Trained models can be evaluated by
python main_eval.py --model_path <MODEL_PATH> --save_path <SAVE_PATH>
which will load all trained models in the <MODEL_PATH> directory and evaluate them in terms of state space divergence and average hellinger distance. The results will be saved to a file in <SAVE_PATH>.
If you find the repository and/or paper helpful for your own research, please cite our work.
@inproceedings{
brenner2025learning,
title={Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data},
author={Manuel Brenner and Elias Weber and Georgia Koppe and Daniel Durstewitz},
booktitle={The Thirteenth International Conference on Learning Representations (ICLR)},
year={2025},
url={https://openreview.net/forum?id=Vp2OAxMs2s}
}
This work was funded by the European Union’s Horizon 2020 programme under grant agreement 945263 (IMMERSE), by the German Ministry for Education & Research (BMBF) within the FEDORA (01EQ2403F) consortium, by the Federal Ministry of Science, Education, and Culture (MWK) of the state of Baden-Württemberg within the AI Health Innovation Cluster Initiative and living lab (grant number 31-7547.223-7/3/2), by the German Research Foundation (DFG) within the collaborative research center TRR-265 (project A06 & B08) and by the Hector-II foundation.