R package for interfacing with Julia's MixedModels library to fit generalized linear mixed-effects models.
Here are some slides from a seminar on how to use jglmm.
Install the package with:
# install.packages("devtools")
devtools::install_github("mikabr/jglmm")
Additionally, you need to have Julia installed, along with the Julia libraries DataFrames.jl, StatsModels.jl, and MixedModels.jl.
The location of your Julia installation needs to be known to the R package, either as the global option JULIA_HOME or the environmental variable JULIA_HOME. For example:
options(JULIA_HOME = "/Applications/Julia-1.2.app/Contents/Resources/julia/bin")
Before using jglmm, you need to do initial setup with jglmm_setup(). It is necessary for every new R session to use the package.
library(jglmm)
jglmm_setup()
To fit a linear regression:
lm1 <- jglmm(Reaction ~ Days + (Days | Subject), lme4::sleepstudy)
To fit a logistic regression:
cbpp <- dplyr::mutate(lme4::cbpp, prop = incidence / size)
gm <- jglmm(prop ~ period + (1 | herd), data = cbpp, family = "binomial",
weights = cbpp$size)
To set the contrasts for a categorical variable:
gm <- jglmm(prop ~ period + (1 | herd), data = cbpp, family = "binomial",
weights = cbpp$size, contrasts = list(period = "effects"))
Access the fixed effects coefficients with tidy(gm) and the fitted response values with augment(gm).
The available response families and their default link functions are:
Bernoulli (LogitLink)
Binomial (LogitLink)
Gamma (InverseLink)
InverseGaussian (InverseSquareLink)
NegativeBinomial (LogLink)
Normal (IdentityLink)
Poisson (LogLink)
Note that the first time you fit a model in a given R session it will take a while, as Julia needs to do some setup operations. Subsequent model fits will be much faster.
For more details on the underlying Julia library, see http://dmbates.github.io/MixedModels.jl/latest/index.html.