ngme2 is an R package used for fitting non-gaussian mixed effects models. These models are fitted using maximum likelihood estimation and preconditioned stochastic gradient descent.
Basic statistical operations such as likelihood evaluations and kriging predictions are also implemented.
Several popular Gaussian random field models can be represented as solutions to stochastic partial differential equations (SPDEs) of the form Lβ(τu) = 𝒲. Here 𝒲 is a Gaussian white noise, L is a second-order differential operator, the fractional power β > 0 determines the smoothness of u. See An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach for further details.
This package aims to address the non-Gaussian extension to the SPDE approach model by replacing the driven noise 𝒲 to be a non-Gaussian noise ℳ. More specificly, a type-G Lévy process.
The increment of a type-G Lévy process can be represented as $$ \gamma + \mu V + \sigma \sqrt{V}Z, $$ where γ, μ are parameters, Z ∼ N(0,1) and is independent of V, and V is a positive infinitely divisible random variable.
One example is the normal inverse Gaussian (NIG) noise. (See vignette("SPDE-approach", package = "ngme2") for more details)
The development version of ngme2 can be installed using the command
remotes::install_github("davidbolin/ngme2", ref = "devel")The stable version of ngme2 can be installed using the command:
install.packages("ngme2", repos = "https://davidbolin.github.io/ngme2/")See also the Installation and Configuration vignette.