Software developed under the project

A software prototype has been developed using the system AD Model Builder. The software prototype is a proof-of-principle, documenting the feasibility of the approach developed under the project Latent variable models handled with optimization aided by automatic differentiation;application to marine resource assessments. The approach has been further developed by Otter Research Ltd., and is available as a separate module for AD Model Builder.

Random effects module for AD Model Builder

AD Model Builder (ADMB) is software package for doing parameter estimation in nonlinear statistical models. ADMB has built into in a powerful function minimizer which employes automatic differentiation. The user of ADMB specifies the model in a template file (tpl-file). The two main components of the tpl-file are:

A specification of the parameters to be estimated.
A description of the objective function for the minimization. In statistics, the objective function will typically be the loglikelihood function (which is maximized rather than minimized).

The Random Effects module allows the user to specify parameters of type "random effects" in addition to the ordinary ADMB parameters (the independent parameters). The random effects are integrated out of the likelihood function by the Laplace approximation, yielding the "marginal likelihood". The independent parameters are then estimated by maximizing the likelihood function with respect to the independent parameters.

Examples

ADMB-RE example collection