tools:::Rd_package_description("glmmrOptim")
glmmrOptim provides algorithms for identifying (approximately) c-optimal experimental designs for experiments described by a generalised linear mixed model.
glmmrOptim provides algorithms for identifying (approximately) c-optimal experimental designs for experiments described by a generalised linear mixed model. Each data row constitutes an observation, which can be grouped into experimental units. The aim is to then find either the discrete subset of experimental units, or the optimal weights on each unit, to minimise the GLS variance criterion. There are four main algorithms:
Reverse greedy search. This combinatorial algorithm starts from the complete set of experimental units and successively removes the unit that minimises the variance until the target sample size is reached.
Local search. This combinatorial algorithm starts from a design of the target sample size and successively swaps units that minimise the variance until no improving swap can be made.
Optimal weights for uncorrelated experimental units. A second-order cone program is used to determine the optimal experimental weights for each unit, where the units are uncorrelated with one another.
Optimal mixed model weights. An algorithm base on the mixed model weights is used to identify the optimal experimental weights, units may be correlated.
The package also provides support for finding the optimal rounding of weights to integers. Robust optimal experimental designs can be identified by including multiple plausible models in the algorithms.
The glmmrOptim package uses the glmmrBase package for model specification and calculation.
The package is still in development and there may still be bugs and errors. While we do not expect the general user interface to change there may be changes to the underlying library as well as new additions and functionality.
tools:::Rd_package_author("glmmrOptim")
Maintainer: tools:::Rd_package_maintainer("glmmrOptim")