placy 0.1.0

PLACy causal discovery for non-stationary time series

PLACy: Robust Causal Discovery in Real-World Time Series with Power-Laws

🏆 ICML 2026 Spotlight (Top 2%)

Matteo Tusoni, Gianmarco Masi, Andrea Coletta, Alessandro Glielmo, Valerio Arrigoni, Nicola Bartolini

Robust Causal Discovery in Real-World Time Series with Power-Laws

Accepted at the International Conference on Machine Learning (ICML 2026) as a Spotlight paper.

📄 Paper: https://arxiv.org/abs/2507.12257

---

Installation

Create the environment:

conda create --name CausalDiscovery python=3.11.11
conda activate CausalDiscovery
pip install -r requirements.txt

---

Quick Start

A complete walkthrough is available in:

📓 example.ipynb

---

Reproducing the Paper Experiments

Run an experiment on synthetic data:

python src/run.py \
    -s SEED \
    --n_vars N_VARS \
    --length LENGTH \
    --causal_strength C \
    --method METHOD \
    --window_length W \
    --stride S

Run an experiment on real data:

python src/run.py \
    -s SEED \
    --dataset DATASET_NAME \
    --method METHOD \
    --window_length W \
    --stride S

---

Citation

If you use PLACy in your research, please cite:

@article{tusoni2025placy,
  title={Robust Causal Discovery in Real-World Time Series with Power-Laws},
  author={Tusoni, Matteo and Masi, Gianmarco and Coletta, Andrea and Glielmo, Alessandro and Arrigoni, Valerio and Bartolini, Nicola},
  journal={International Conference on Machine Learning (ICML)},
  year={2026}
}