flopsearch 0.2.0

Python package providing an implementation of the FLOP causal discovery algorithm for linear additive noise models

flopsearch

Python package providing an implementation of the FLOP causal discovery algorithm for linear additive noise models.

Installation

flopsearch can be installed via pip:

pip install flopsearch

Citing FLOP

If you use FLOP in your scientific work, please cite this paper:

@article{cifly2025,
  author  = {Marcel Wien{"{o}}bst and Leonard Henckel and Sebastian Weichwald},
  title   = {{Embracing Discrete Search: A Reasonable Approach to Causal Structure Learning}},
  journal = {{arXiv preprint arXiv:2510.04970}},
  year    = {2025}
}

Example

A simple example run of the FLOP algorithm provided by flopsearch.

import flopsearch
import numpy as np
from scipy import linalg

p = 10
W = np.diag(np.ones(p - 1), 1)
X = np.random.randn(10000, p).dot(linalg.inv(np.eye(p) - W))
X_std = (X - np.mean(X, axis=0)) / np.std(X, axis=0)
flopsearch.flop(X_std, 2.0, restarts=50)

Input and Output

As input, FLOP takes the data matrix, the BIC penalty parameter (we recommend 2.0 as a default choice) and either a timeout (in seconds) or the number of ILS restarts to control how long the search runs.

The output of FLOP is a CPDAG encoded with an adjacency matrix whose entry in row i and column j is 1 in case of a directed edge from the i-th to the j-th variable and 2 in case of an undirected edge between those variables (in case of an undirected edge, the entry in row j and column i is also 2, that is each undirected edge induces two 2's in the matrix).