Jupyter kernel uses different Python version than Conda environment - jupyter-notebook

Question

I am not able to link Jupyter kernels to their parent Conda environments. After creating a new kernel linked to Conda environment, I'm getting a different version of Python and its dependencies inside Jupyter lab.
Here are the steps I followed:
Created a conda environment using:
conda create -n nlp python=3.6
conda activate nlp
(nlp) ➜ ~ python --version
Python 3.6.9 :: Anaconda, Inc.
(nlp) ➜ ~ which python
/anaconda3/envs/nlp/bin/python
Inside the environment I created a Jupyter kernel with:
(nlp) ➜ ~ python -m ipykernel install --user --name=nlp
Installed kernelspec nlp in /Users//Library/Jupyter/kernels/nlp
Investigating the created json file for the kernel:
(nlp) ➜ ~ cat /Users/<username>/Library/Jupyter/kernels/nlp/kernel.json
{
"argv": [
"/anaconda3/envs/nlp/bin/python",
"-m",
"ipykernel_launcher",
"-f",
"{connection_file}"
],
"display_name": "nlp",
"language": "python"
}%
It seems to be pointing to the environment version of Python
But when I start Jupyter Lab and select the nlp kernel, I get a different version of Python and some dependencies are missing
!python --version
Python 3.5.6 :: Anaconda, Inc.
!which python
/anaconda3/bin/python

Answer 1

Could you please try the following steps:
conda activate nlp
conda install ipykernel
ipython kernel install --name nlp --user
After these steps please try changing the kernel again in jupyter lab to "nlp".
Thanks.

Answer 2

this behavior is actually normal in Jupyter lab.
If you run
import sys
print(sys.version)
!python --version
in a notebook, the print statement will give you the Python version of the conda env, while the second will give you the Python version of your base env.
The easiest workaround for this is to simply pip install jupyterlab in your conda env and then run jupyter lab in your conda env. Then, there will not be a mismatch in Python versions between the new "base" env and the conda env which will help clear up any DLL problems.
It's probably not best practice, but you do what you gotta do when working with legacy code, ig.

Answer 3

Can you try this :
# in base env
conda install nb_conda_kernels
conda activate nlp
conda install ipykernel
conda install ipywidgets
# install kernelspec
python -m ipykernel install --user --name nlp --display-name "nlp env"
When you run jupyter notebook, you will see 2 nlp kernels. Use the one with "Python [conda:env:nlp]"