The paper proposes a novel constraint-based algorithm for inferring causal networks that automatically determines topological thresholds from the data. Two methods are presented for determining the threshold:
The Connected method seeks a set of edges that leaves no disconnected nodes in the network.
The Knee method seeks the largest connected component in the data, finding the point of greatest curvature in the size of the largest component vs ranked edges.
The algorithm uses these thresholds as constraints to prune the network, removing statistically weak edges. It is tested on both synthetic and real-world networks, and compared to the benchmark PC algorithm.
The results show that the proposed algorithm is generally faster and more accurate than the PC algorithm, especially for larger networks. A novel asymmetric measure called Net Influence is also introduced, which allows the algorithm to directly infer the directionality of edges.
The key advantages of the algorithm are its ability to automatically determine appropriate thresholds, its computational efficiency, and its performance in accurately inferring causal networks from discrete data.
In un'altra lingua
dal contenuto originale
arxiv.org
Approfondimenti chiave tratti da
by Filipe Barro... alle arxiv.org 04-24-2024
https://arxiv.org/pdf/2404.14460.pdfDomande più approfondite