This repository implements paper "A Bregman Method for Structure Learning on Sparse Directed Acyclic Graphs" by Manon Romain and Alexandre d'Aspremont.
Run pip install causal-lasso from Terminal.
Run git clone https://github.com/manon643/causal_lasso.git. Install dependencies though pip install -r docs/requirements.txt.
For now, the default solver used at each iteration is Mosek. We plan to provide an open source implementation in the near future.
MOSEK's license is free for academic use, first obtain your license here using institutional email and place the obtained file mosek.lic in a file called:
%USERPROFILE%\mosek\mosek.lic (Windows)
$HOME/mosek/mosek.lic (Linux, MacOS)
Further information available here.
Minimal testing code is:
import numpy as np
import networkx as nx
from causal_lasso.solver import CLSolver
X = np.random.random((1000, 30)) # Replace with your data
lasso = CLSolver()
W_est = lasso.fit(X)
nx.draw(nx.DiGraph(W_est))
A more detailed tutorial with synthetic data is available in examples/tutorial.ipynb.
If you want to apply the algorithm to your own data, you should check out examples/real_datasets.ipynb