This package implements tools for integrating out (marginalizing) parameters from log-probability functions, such as likelihoods and posteriors of statistical models. The functionality is similar to that of the TMB (Template Model Builder) package for R, but implemented in pure Julia.
This approach is useful for fitting models that incorporate random effects or latent variables that are important for structuring the model, but whose exact values may not be of interest in and of themselves. In the language of regression, these are known as "random effects," and in other contexts, they may be called "nuisance parameters." Whatever they are called, we hope that our log-density function can be optimized faster and/or more reliably if we focus on the parameters we are actually interested in, averaging the log-density over all possible values of the ones we are not. Read more in the docs.