glmmTMB

glmmTMB is an R package for fitting generalized linear mixed models (GLMMs) and extensions, built on Template Model Builder, which is in turn built on CppAD and Eigen. It is intended to handle a wide range of statistical distributions (Gaussian, Poisson, binomial, negative binomial, Beta …) and zero-inflation. Fixed and random effects models can be specified for the conditional and zero-inflated components of the model, as well as fixed effects for the dispersion parameter.

Where to ask questions

the r-sig-mixed-models mailing list for general questions about glmmTMB usage and mixed models (please subscribe to the list before posting)
the glmmTMB issues list for bug, infelicity, and wishlist reporting
the TMB users forum for TMB-specific questions
maintainer e-mail only for urgent/private communications

Please do not cross-post, i.e. ask the same question in more than one forum, unless it’s suggested that you have posted in the wrong place, or unless you receive total silence in one forum. In the latter case it may be better to send a reminder/“bump” message to the original forum; in any case you should mention in your new message where/when you’ve previously asked the question.

See here for the development site.

Installation

Simple cases

from CRAN (release version): install.packages("glmmTMB"). (On Windows and MacOS this will install binary packages, by default: see below for installation from source and why you might want to do that.)
from GitHub (development version, from source): use install.packages() to install the TMB and remotes packages from CRAN, then remotes::install_github("glmmTMB/glmmTMB/glmmTMB"). If the install fails at the vignette-building step, try specifying build_vignettes=FALSE within the install_github call. You will need to have development tools (compilers etc.) installed: see here or here. (As well as being more up-to-date, the development version may contain new bugs or untested features!)
from GitHub (development version, from binary): A binary release of the development version may be available for your operating system/version of R from the glmmTMB repository here. If a sufficient version isn’t available and you are having insurmountable problems installing the current version of the package from source yourself, please contact the maintainers.

Complications

glmmTMB/TMB/Matrix mismatches

glmmTMB should be run on a system with the same version of TMB installed that was originally used to build the package; TMB and Matrix have a similar dependency. If you update your system with binary versions of the TMB or Matrix packages from CRAN that are newer than the version of glmmTMB on CRAN, you’ll get warnings or error messages.

You can:

re-install the entire Matrix > TMB > glmmTMB stack from source (this is slight overkill, you might not need to re-install the whole stack, but it doesn’t hurt):
- Make sure that you have development tools installed. See here for MacOS and here for Windows (if you’re on Linux, you probably know what you’re doing with this …) You will need both C++ and Fortran compilers (gfortran).
- start a clean R/RStudio session (make sure no packages other than the base/core packages are loaded)
- install.packages("Matrix"): this should install the latest version of Matrix
- install.packages("TMB", type = "source") - this installs the latest version of TMB in a way that is binary-compatible with the latest Matrix
- install.packages("glmmTMB", type = "source") - this installs the latest version of glmmTMB in a way that is binary-compatible with the latest TMB
hope that updated binary versions are available here for your OS and R version. You can check http://glmmtmb.github.io/glmmTMB/repos/bin/[OS]/contrib/[R_version]/PACKAGES, where [OS] is “macosx” or “windows”, and [R_version] is the major version of R you’re using (e.g. 4.1). See “install development version from GitHub, binary” above. (Windows and MacOS binaries of TMB built with newer versions of the Matrix package may be available here as well. Try: install.packages("TMB", repos="https://glmmTMB.github.io/glmmTMB/repos").)
Use the groundhog package to install binary versions of Matrix, TMB, and glmmTMB from a date when they were consistent with each other:

(note: this recipe is not well tested … if you try it and encounter problems, please post an issue)

## load (installing if necessary) the groundhog package
while (!require("groundhog", quietly=TRUE)) install.packages("groundhog")
## retrieve build date of installed version of glmmTMB
bd1 <- as.character(asDateBuilt(packageDescription("glmmTMB",fields="Built")))
groundhog.library("TMB", bd1)
bd2 <- as.character(asDateBuilt(packageDescription("TMB",fields="Built")))
groundhog.library("Matrix", bd2)

The only disadvantage to this approach is that your versions of TMB and Matrix will be behind the version on CRAN; you might be missing out on some bug fixes or improvements, and eventually you may find that updates of other packages require newer versions of these packages. (If you accidentally update the packages from CRAN, you’ll have to redo this step.) - Install older versions of TMB and/or Matrix from CRAN, from source using remotes::install_version("[pkg]","[xxxx]"), where [pkg] is TMB or Matrix and [xxxx] is the older package version referred to in the first error message you received.

OpenMP compilation

This section and the next may become outdated as new versions of operating systems are released; please notify the maintainers if you run into trouble.

glmmTMB enables parallel (threaded) computations based on OpenMP (see the parallel vignette for more information). OpenMP will be available automatically if your system supports it, but this may depend on the details of your operating system, compiler, compilation flags used when your R executable was built, etc.; in particular, see here for tips on enabling OpenMP for older (<= El Capitan/10.11.4) MacOS systems. (That page suggests using optimization level -O3, which may cause problems for glmmTMB.)

The maximum number of threads used defaults to 48; to increase this value when installing from source, you can use

withr::with_makevars(c(PKG_CPPFLAGS="-DCPPAD_MAX_NUM_THREADS=128"), {
  remotes::install_github("glmmTMB/glmmTMB/glmmTMB")
}, assignment="+="
)

(or the equivalent for remotes::install_cran()).

MacOS compilation issues

if you get errors about library not found for -lgfortran or library not found for -lquadmath you may need to follow these instructions to update your Fortran compilers.
if you get errors about unsupported option '-fopenmp'you need to turn off OpenMP compilation by adding the line SHLIB_OPENMP_CFLAGS= to your ~/.R/Makevars file. If you’ve already done step #1 this file will already exist; use a text editor to add the line above. Otherwise, you need to create it.

After updating your compilers (if necessary) and turning off OpenMP compilation, re-try the installation from source.