Maximizing the expected log posterior density (ELPD) is a common objective function in Bayesian inference and Bayesian neural networks, and I'd be interested in having it as a loss function for a package I'm building. I'd be happy to implement it if someone around here is willing to help me figure out the API (I haven't used JuliaML before; I'm building a package that does efficient LOO-CV for Bayesian models).
(Note: ELPD is the generalization of log-loss/cross-entropy loss to regression, rather than classification, problems.)
Maximizing the expected log posterior density (ELPD) is a common objective function in Bayesian inference and Bayesian neural networks, and I'd be interested in having it as a loss function for a package I'm building. I'd be happy to implement it if someone around here is willing to help me figure out the API (I haven't used JuliaML before; I'm building a package that does efficient LOO-CV for Bayesian models).
(Note: ELPD is the generalization of log-loss/cross-entropy loss to regression, rather than classification, problems.)