I have a conda environment which I made available as a kernel to my Jupyter instance by running:
python -m ipykernel install --user --name my-env-name --display-name "Python (my-env-name)"
With this environment I wanted to use R in Jupyter taking advantage of rpy2's %load_ext rpy2.ipython command to enable the %%R magic. However, rpy2 is employing my global R and not the one installed in my conda environment. I checked my R home via:
%%R
R.home()
(I can also check the situation with %run -m rpy2.situation in Jupyter notebook (source), however this seems to be broken in rpy2 somewhere between versions 3.1.0 and 3.2.1 ... at least for me it was throwing UnboundLocalError: local variable 'rpy2' referenced before assignment in 3.1.0 and it was working for 3.2.1).
How can I make my Jupyter notebook load the R installation from my conda environment?
There are two approaches to solve this, a local (for individual Jupyter notebooks) and a global one (for the kernel itself). Both are related to setting the R_HOME environment variable.
Local (source):
Before calling %load_ext rpy2.ipython in your Jupyter notebook, run:
import os
os.environ['R_HOME'] = '/home/your/anaconda3/envs/myenv/lib/R' #path to your R installation
Global:
Find your kernel directory via: jupyter kernelspec list and edit the file kernel.json. Update the JSON by adding:
"env": {"R_HOME":"/home/your/anaconda3/envs/my-env-name/lib/R"}, then restart your kernel (you might have to restart Jupyter as well).
LD_LIBRARY_PATH)Recently, I tried running rpy2 in jupyter again after setting up a new environment using conda:
conda config --add channels conda-forge
conda config --set channel_priority strict
conda create -n myenv python=3.7
conda activate myenv
conda install r-essentials pandas rpy2
And this time I ran into the following issue when trying to either %load_ext rpy2.ipython (Jupyter) or simply import rpy2.robjects (any script):
>>> import rpy2.robjects
Warning message:
package ‘methods’ was built under R version 3.6.3
Error: package or namespace load failed for ‘stats’ in dyn.load(file, DLLpath = DLLpath, ...):
unable to load shared object '/home/your/anaconda3/envs/myenv/lib/R/library/stats/libs/stats.so':
/home/your/anaconda3/envs/myenv/lib/R/library/stats/libs/stats.so: undefined symbol: MARK_NOT_MUTABLE
During startup - Warning messages:
1: package ‘datasets’ was built under R version 3.6.3
2: package ‘utils’ was built under R version 3.6.3
3: package ‘grDevices’ was built under R version 3.6.3
4: package ‘graphics’ was built under R version 3.6.3
5: package ‘stats’ was built under R version 3.6.3
6: package ‘stats’ in options("defaultPackages") was not found
R[write to console]: Error: package or namespace load failed for ‘tools’ in dyn.load(file, DLLpath = DLLpath, ...):
unable to load shared object '/home/your/anaconda3/envs/myenv/lib/R/library/tools/libs/tools.so':
/home/your/anaconda3/envs/myenv/lib/R/library/tools/libs/tools.so: undefined symbol: R_NewPreciousMSet
R[write to console]: Error in dyn.load(file, DLLpath = DLLpath, ...) :
unable to load shared object '/home/your/anaconda3/envs/myenv/lib/R/library/tools/libs/tools.so':
/home/your/anaconda3/envs/myenv/lib/R/library/tools/libs/tools.so: undefined symbol: R_NewPreciousMSet
R[write to console]: In addition:
R[write to console]: Warning message:
R[write to console]: package ‘tools’ was built under R version 3.6.3
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/your/anaconda3/envs/myenv/lib/python3.7/site-packages/rpy2/robjects/__init__.py", line 20, in <module>
import rpy2.robjects.functions
File "/home/your/anaconda3/envs/myenv/lib/python3.7/site-packages/rpy2/robjects/functions.py", line 12, in <module>
from rpy2.robjects import help
File "/home/your/anaconda3/envs/myenv/lib/python3.7/site-packages/rpy2/robjects/help.py", line 43, in <module>
tools_ns = _get_namespace(StrSexpVector(('tools',)))
File "/home/your/anaconda3/envs/myenv/lib/python3.7/site-packages/rpy2/rinterface_lib/conversion.py", line 44, in _
cdata = function(*args, **kwargs)
File "/home/your/anaconda3/envs/myenv/lib/python3.7/site-packages/rpy2/rinterface.py", line 621, in __call__
raise embedded.RRuntimeError(_rinterface._geterrmessage())
rpy2.rinterface_lib.embedded.RRuntimeError: Error in dyn.load(file, DLLpath = DLLpath, ...) :
unable to load shared object '/home/your/anaconda3/envs/myenv/lib/R/library/tools/libs/tools.so':
/home/your/anaconda3/envs/myenv/lib/R/library/tools/libs/tools.so: undefined symbol: R_NewPreciousMSet
The issue seemed to have been a screwed up R "situation" (check via %run -m rpy2.situation in Jupyter or simply python -m rpy2.situation on the command line), which had R's additions to LD_LIBRARY_PATH: pointing to and old, globally installed R version.
I had to manually unset the LD_LIBRARY_PATH to solve this issue. This path can be set / unset analogously to R_HOME.
PS: I found R_HOME and LD_LIBRARY_PATH set in my .bashrc to custom (from source) R installation. Which confused the Jupyter kernel obviously. Not smart ;)
PPS: rpy2.situation still tells me that there is a Warning: The environment variable R_HOME differs from the default R in the PATH.:
Looking for R's HOME:
Environment variable R_HOME: /home/your/anaconda3/envs/myenv/lib/R
Calling `R RHOME`: /home/your/anaconda3/envs/jupyter-env/lib/R
Environment variable R_LIBS_USER: None
Warning: The environment variable R_HOME differs from the default R in the PATH.
Which worries me that R actually defaults to the R installed for the Jupyter installation. So if anybody has comments about this, I would be grateful.