Jupyter notebook kernel different from conda environment

Question

I'm encountering problems while trying to reproduce a conda environment in a jupyter notebook kernel. I've created a conda environment, mlflow, and by activating it, I can import mlflow, as you can see below:

[ 12:36:18 ] ~ base ❯ conda activate mlflow
[ 12:36:21 ] ~ mlflow ❯ python
Python 3.7.0 (default, Jun 28 2018, 08:04:48) [MSC v.1912 64 bit (AMD64)] :: Anaconda, Inc. on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import mlflow
>>>

Then I proceed to install the kernel in jupyter, as:

python -m ipykernel install --name mlflow

However, once in the mlflow kernel in jupyter, I cannot import the same module, mlflow. Why could that be? It suspect that the issue is that C:\Users\userx\AppData\Roaming\jupyter\kernels\mlflow\kernel.json is pointing to the wrong python:

{
 "argv": [
  "C:\\Users\\userx\\Anaconda3\\python.exe",
  "-m",
  "ipykernel_launcher",
  "-f",
  "{connection_file}"
 ],
 "display_name": "mlflow",
 "language": "python"
}

Whereas it should be pointing to the python in the mlflow environment. However, I tried to change it and the caused the notebook to fail when launching.

More details:

[ 12:39:27 ] ~ base ❯ jupyter --path
config:
    C:\Users\userx\.jupyter
    C:\Users\userx\Anaconda3\etc\jupyter
    C:\ProgramData\jupyter
data:
    C:\Users\userx\AppData\Roaming\jupyter
    C:\Users\userx\Anaconda3\share\jupyter
    C:\ProgramData\jupyter
runtime:
    C:\Users\userx\AppData\Roaming\jupyter\runtime

EDIT

It seems to work if I install jupyter in the environment, and launch a notebook from the environment. However, it'd still be nice to know how to do the same using the jupyter in base.

If Jupyter is installed through Conda, e.g., in your **base**, then I find the cleaner workflow is not to manually register envs, but instead install `nb_conda_kernels` in the environment with `jupyter` and it will automatically discover any environments with appropriate kernel packages installed (e.g., `ipykernel`, `r-irkernel`). See [this related answer](https://stackoverflow.com/a/43197286/570918). — merv, Dec 01 '20 at 14:55

score 1 · Accepted Answer · edited Dec 01 '20 at 12:36

1

Try giving the full conda python path and see if that resolves it:

C:\Users\<username>\anaconda\envs\<environment-name>\bin\python -m ipykernel install --name mlflow

The reason ipykernel was not found was because it was not installed in the right python environment. Using "python -m" may reference the native python outside conda or maybe the python in the base environment. Specifying the full path just ensures that it gets installed in the right environment.

edited Dec 01 '20 at 12:36

yatu

86,083
12
84
139

answered Dec 01 '20 at 11:55

Swishonary

93
1
6

1

That indeed seems to do the trick! So could you add more details? So the issue was that the kernel is defined to launch the wrong python? – yatu Dec 01 '20 at 12:03
I have edited the answer to provide more info. Please do 'mark as the answer' if this resolves the issue. Thanks. – Swishonary Dec 01 '20 at 12:36
1

Thanks,very useful! – yatu Dec 01 '20 at 12:36

Jupyter notebook kernel different from conda environment

1 Answers1