arXiv:2407.00644v1 Announce Kind: new
Summary: Graphical fashions function efficient instruments for visualizing conditional dependencies between variables. Nonetheless, because the variety of variables grows, interpretation turns into more and more troublesome, and estimation uncertainty will increase as a result of giant variety of parameters relative to the variety of observations. To handle these challenges, we introduce the Clusterpath estimator of the Gaussian Graphical Mannequin (CGGM) that encourages variable clustering within the graphical mannequin in a data-driven means. By way of using a clusterpath penalty, we group variables collectively, which in flip ends in a block-structured precision matrix whose block construction stays preserved within the covariance matrix. We current a computationally environment friendly implementation of the CGGM estimator through the use of a cyclic block coordinate descent algorithm. In simulations, we present that CGGM not solely matches, however oftentimes outperforms different state-of-the-art strategies for variable clustering in graphical fashions. We additionally reveal CGGM’s sensible benefits and flexibility on a various assortment of empirical purposes.
Supply hyperlink