CausalDisco 0.2.0

Baseline algorithms and analytics tools for Causal Discovery.

CausalDisco 🪩

CausalDisco contains baseline algorithms and analytics tools for Causal Discovery. The package can be installed by running pip install CausalDisco. Additional information can be found in the documentation.

Baseline Algorithms

Find the following baseline algorithms in CausalDisco/baselines.py

• R²-SortnRegress
• Var-SortnRegress
• Random-SortnRegress

Analytics Tools

Find the following analytics tools in CausalDisco/analytics.py

• R²-sortability
• Var-sortability
• order_alignment

Sources

If you find our algorithms useful please consider citing

• Beware of the Simulated DAG!
• Simple Sorting Criteria Help Find the Causal Order in Additive Noise Models.

@article{reisach2021beware,
  title={Beware of the Simulated DAG! Causal Discovery Benchmarks May Be Easy to Game},
  author={Reisach, Alexander G. and Seiler, Christof and Weichwald, Sebastian},
  journal={Advances in Neural Information Processing Systems},
  volume={34},
  year={2021}
}

@article{reisach2023simple,
  title={Simple Sorting Criteria Help Find the Causal Order in Additive Noise Models},
  author={Reisach, Alexander G. and Tami, Myriam and Seiler, Christof and Chambaz, Antoine and Weichwald, Sebastian},
  journal={arXiv preprint arXiv:2303.18211},
  year={2023}
}

A Simple Example

# --- sample data from a linear SCM:

import numpy as np
from scipy import linalg
from CausalDisco.analytics import var_sortability, r2_sortability

d = 10
W = np.diag(np.ones(d-1), 1)
X = np.random.randn(10000, d).dot(linalg.inv(np.eye(d) - W))
X_std = (X - np.mean(X, axis=0))/np.std(X, axis=0)

# --- run analytics and print results:

from CausalDisco.analytics import (
    var_sortability,
    r2_sortability,
    snr_sortability
)

print(
    f'True\n{W}\n'
    f'var-sortability={var_sortability(X, W):.2f}\n'
    f'R^2-sortability={r2_sortability(X, W):.2f}\n'
    f'SNR-sortability={snr_sortability(X, W):.2f}'
)

# --- run baselines and print results:

from CausalDisco.baselines import (
    random_sort_regress,
    var_sort_regress,
    r2_sort_regress
)

print(
    f'True\n{W}\n'
    '--- randomRegress ---\n'
    f'Recovered:\n{1.0*(random_sort_regress(X)!=0)}\n'
    f'Recovered standardized:\n{1.0*(random_sort_regress(X_std)!=0)}\n'
    '--- varSortnRegress ---\n'
    f'Recovered:\n{1.0*(var_sort_regress(X)!=0)}\n'
    f'Recovered standardized:\n{1.0*(var_sort_regress(X_std)!=0)}\n'
    '--- r2SortnRegress ---\n'
    f'Recovered:\n{1.0*(r2_sort_regress(X)!=0)}\n'
    f'Recovered standardized:\n{1.0*(r2_sort_regress(X_std)!=0)}\n'
)