Would anyone have instructions how to set up Anaconda and MNE-Python to
run on a linux CentOS 6.5 workstation (non-Martinos)? My system possibly
already has some anaconda components (= /usr/lib/anaconda and
/usr/lib/anaconda/runtime/ exist). Typing "anaconda" in a tcsh terminal
while logged in as root launches an anaconda GUI installer.
I plan to install the developmental version of mne-python. I am using
tcsh. I can do this as root.
However, I am unsure from which dir I should do the installations, and how
to set the paths correctly so that MNE-python will talk with Anaconda
(instead of other python versions included with CentOS).
Since several users in my lab will be using Anaconda and MNE-python on
this same linux box, would it make sense to have one central anaconda
installation instead of installing separately for every user? (I hope to
avoid having multiple python versions and virtualenv:s). If so, how would
the instructions change?
Eventually I would also like to install NVIDIA CUDA...
If `which pip` points to a `pip` within the Anaconda directory, then it
indicates you have the anaconda environment set up properly, and that doing
commands with `pip` will be reflected when you use the Anaconda python (and
not the system Python). I'd look at standard Anaconda tutorials to figure
out how to get that set up properly.
It doesn't really matter which directory you run that `pip` command from,
they all end up installing the package to the same place. After it's done,
you should be able to delete whatever temporary directory it created/used
during installation.
I don't know how you got a /usr/lib/anaconda -- it doesn't exist on Ubuntu
and my experience with CentOS is limited, so I'm not sure if it's part of
the system (I doubt it), or if someone installed that manually (most
likely). I'd look into Anaconda tutorials to see if someone has that as a
set of instructions, or if it's standard for CentOS or some CentOS package.
And you can always try using it, and if it doesn't work, set up your own
other Anaconda environment.
Regarding system-wide vs user-specific installs -- it's easy to install
things system-wide when using a system python by omitting the `--user`
option during install. I have a feeling that since your anaconda is in
`/usr/lib/anaconda`, omitting the `--user` option will similarly install it
system-wide for Anaconda. Generally the `--user` option installs a package
to a place where only the given user has access (e.g.,
~/.local/lib/python2.7/site-packages/ for a system python), whereas the
standard install puts it in a system location that anyone can access.
Having one central version is fine. Users can always install their own
newer / developmental version using the `--user` option if they wanted to
e.g., contribute new features or get the latest updates. The user paths
take precedence over the system ones in the Python path-searching hierarchy.
PyCUDA is easiest to install on Ubuntu 14.04+, where it can be installed by
apt-get / software repos. I assume you don't want to do an OS reinstall,
though PyCUDA can be a bit tough to get installed properly manually, but
it's possible. You'll have to install the NVIDIA CUDA toolkit (unless
there's a recent-enough yum version, but I doubt it), ensure the paths are
set properly for it, and then follow a tutorial on how to build PyCUDA. I
had to do it once or twice on earlier versions of Ubuntu that didn't have
it in synaptic and it was doable, but took a bit of perseverance.
PyCUDA can be a bit tough to get installed properly manually, but