Online Program

Dimension Reduction and Computational Time (309928)

Keywords: R code, Dimension reduction, computational time, GLMMs

The R package glmm (Knudson) allows users to fit generalized linear mixed models (GLMMs). Despite the package’s functionality, glmm is computationally expensive due to the millions of calculations involved with Monte Carlo Likelihood Approximation. The R package glmm has already been optimized with parallel computing using clusters but more research was needed to find additional ways to reduce the amount of computational time through reducing the dimensions of the random effects. The methods that were initially explored included rounding and quantile-grouping. Of these methods it was determined that quantile-grouping provided reduced computational time while maintaining predictive integrity. This was measured by comparing computational time and prediction outputs of the glmm original computations with that of different methods of reducing dimensions. Our research shows that reducing the dimensions of the random effects is an effective way of lessening computational time while maintaining accurate fits for generalized linear mixed models.