Sporadic CCM

This package contains the python implementation of the paper : "Inferring Causal Dependencies between Chaotic Dynamical Systemsfrom Sporadic Time Series".

Authors : Edward De Brouwer, Adam Arany, Jaak Simm and Yves Moreau.

Dependencies.

gru_ode : https://github.com/edebrouwer/gru_ode_bayes

skccm : https://skccm.readthedocs.io/en/latest/

Installation.

From the top directory, run :

pip install -e . 

Running Code

Data Generation

Generation of sporadic double pendulum trajectories is computed using :

python data_generation_script.py

Filtering of the sporadic time series

The following script will train a GRU-ODE-Bayes filtering model on top of the given data and reconstruct the full trajectory accordingly.

.\launch_gru_ode.sh

Trained models are saved in the trained_models folder.

Causal direction inference

We can then compute the scores for causal dependence between dynamic systems :

python gruode_scores.py

The scores are saved in results_ccm.csv