Unable to change conda environment in kernel

[iguy0]

Hi,

I'm trying to add/change the conda/python environment and i'm unable to do so.

Steps to reproduce:

docker run --gpus all -d -it -p 8848:8888 -v /home/USER:/home/jovyan/work -e GRANT_SUDO=yes -e JUPYTER_ENABLE_LAB=yes --user root cschranz/gpu-jupyter:v1.5_cuda-11.6_ubuntu-20.04_python-only

Log into JupyterLab and add a different kernel/conda environment using the terminal

mkdir /home/jovyan/envs
conda init bash
exec $SHELL
conda create -p /home/jovyan/envs/python3.11 python=3.11
conda activate /home/jovyan/envs/python3.11
pip install --user ipykernel
python -m ipykernel install --user --name=python3.11
conda deactivate

at this point the kernel is available, but selecting the kernel gives me this

Installing packages via pip and conda end up under /opt/conda when importing them uses the 3.11 environment, and gives a ImportError error. How to fix this ?

[iguy0]

I got it going. The solution and additional useful commands are below:

Create environment paths and init conda - this is executed from Terminal in Jupyterlab

mkdir /home/jovyan/envs
conda init bash
exec $SHELL

Install in conda base environment

conda install nb_conda_kernels # https://github.com/Anaconda-Platform/nb_conda_kernels

Install new Environment and kernel

conda create -p /home/jovyan/envs/python3.11 python=3.11.6
conda activate /home/jovyan/envs/python3.11
conda install ipykernel
sudo ipython kernel install --name=python3.11 --display-name "Python 3.11 (Darts)"
conda deactivate

Restart container to recognize new kernel

Remove kernel

jupyter kernelspec list
jupyter kernelspec uninstall unwanted-kernel

[iguy0]

Not a solution for the original issue but allows me to create additional environments/kernels

[benz0li]

@iguy0 See jupyter/docker-stacks#2050.

FYI @mathematicalmichael

[ChristophSchranz]

Closing as there were no contributions for a while and it seems there is a workaround.