Learning the Dynamics of Sparsely Observed Interacting Systems (arXiv)

Requirements

For reproductibility purposes, we advise to install the project in a dedicated virtual environment to make sure the specific requirements are satisfied. Recommended Python version: 3.7.x.

To install requirements, use

pip install -r requirements.txt

Then install our code as a package using

pip install -e .

Running an experiment

This code is based on the sacred framework. To run an experiment, you need to specify a configuration in experiments/configs.py and then run

python experiments/simulation_main.py name_config

For example, to reproduce the experiment of the paper where we vary the number of sampling points of the target time series (Figure 2), run

python experiments/simulation_main.py y_sampling

The results of the experiments are saved in a folder. The notebook notebooks/Analyse results.ipynb illustrates how to then load the results from the experiment files.

See also the notebook notebooks/Demo models.ipynb for some simple examples.

Reproducing the paper experiments

All hyperparameters used in the article are given in the configuration file experiments/configs.py.

Toy experiment

Take a look at notebooks/Toy model.ipynb.

HGR forecasting during the COVID-19 pandemic

Please refer to the notebook notebooks/Covid preprocessing.ipynb for details on the pre-processing of the COVID-19 data and to notebooks/Covid results plots.ipynb for result analysis. The original data can be found here.