Installing cmdstanR on Windows
Jordan Hart, Joe Wilde
Source:vignettes/windows_installation.Rmd
windows_installation.Rmd
Introduction
The installation of cmdstanR can be troublesome on Windows computers, even more so for networked computers. Don’t let this put you off though, once you get cmdstanR installed you’ll be able to run bisonR and many other packages at speed and with less buggy-ness. In this article we’ve compiled a list of problems and fixes that have worked for us. In some cases we’ve resorted to editing the registry to fix some of the problems, but be careful when doing this and make sure only to make changes if you’re confident, as the registry holds a lot of important information used by the OS. Don’t change anything if you don’t have a complete backup or don’t fully understand what is being changed.
Fixes
Installing into a network directory/issues with tar.exe
If you’re using a computer on a network (i.e. in most universities
and businesses), there’s a good chance that your HOME
environment variable will be a network location. This will cause many
problems with installation and even if it did work, would likely be much
slower too. The easiest solution to this problem is to create a local
directory (i.e. somewhere in your C drive that isn’t indexed by network
backups etc) called “cmdstan”. This will be the install location for
cmdstan, and you can specify this directory when using
install_cmdstan()
:
install_cmdstan(dir="C:/path/to/directory/cmdstan")
RTools 4.x was not found but is required to run CmdStan with R version 4.x.
This problem occurs because an environment variable was not set
properly when Rtools was installed. This appears to be a common problem,
but it can be fixed by adding an environment variable
RTOOLS40_HOME
(for Rtools 4.0) or
RTOOLS42_HOME
(for Rtools 4.2). The environment variable
should point to the location of your Rtools installation.
Supplied csv file is corrupt
This occurs after running the sampler. We don’t know the source of
this problem, but installing the github version of cmdstanR
seems to fix the problem. This can be done by running the following
code:
remotes::install_github("stan-dev/cmdstanr")
If this doesn’t work, make sure you’re using the latest version of R (R 4.2.1) at the time of writing.
Error: CmdStan path has not been set yet.
This can happen on the R session following the session in which you
first installed cmdstanR. For some reason the environment variable
CMDSTAN
, which was set on installation, has been
overwritten. We think this may be due to the configuration of managed
computers. The solution to this problem is to set CMDSTAN
to the path of your cmdstan installation. Specifically, this needs to
point to the installation version directory
(e.g. .cmdstan/cmdstan-2.30.0
), rather than just the
.cmdstan/
or cmdstan/
directory.