Why does the virtual environment in Jupyterlab has access to packages that were not installed?

My question is why I can access pandas in a jupyter notebook even though it was not installed in the used environment.

I created a virtual environment:
python -m venv ml_skt_book
I then activated it:
ml_skt_book\Scripts\activate.bat
I then installed numpy:
pip install numpy
I have three packages now: numpy, pip and setuptools

I then link it to the jupyter notebook kernel:
python -m ipykernel install --name=ml_skt_book

That worked. I then checked in the command line if numpy works and pandas not. This was successfull.

I then opened jupyterlabs:
jupyter-lab
started a new notebook with the environment "ml_skt_book". That worked.

I then checked the packages:

On the right you can see that the "ml_skt_book" environment is used. It therefore doesn't make sense that pandas can be imported.

I think the venv has access to the packages of its 'base' python by default unless you disable it by setting --system-site-packages to false.

See the doc for details.

Edit Mar 1st 2023

I finally figured out how this whole thing works.

Once you run python -m ipykernel install --name=ml_skt_book, it actually generate a kernelspec folder named ml_skt_book in your home path or jupyter path (the console output tells when you run that). If you forgot, by default it should be in C:\ProgramData\jupyter\kernels (windows) or ~/.local/share/jupyter/kernels (linux). Inside the folder there is a kernel.json which looks like

{
 "argv": [
  "/usr/bin/python",
  "-m",
  "ipykernel_launcher",
  "-f",
  "{connection_file}"
 ],
 "display_name": "ml_skt_book",
 "language": "python",
 "metadata": {
  "debugger": true
 }

and that python in argv is the python used for the kernel.

So you'll need to check what's in your kernel.json and see if it points to the python executable in your ml_skt_book venv, or to the base environment.

According to my experiment, you'll need to run python -m ipykernel install --name=ml_skt_book with your venv ml_skt_book activated to generate kernelspec with your venv python in argv. If you run that command under base environment, it generates kernelspec with name "ml_skt_book" only, but the python executable still the one in base env, that is to say, meaningless.

Also, if you didn't have ipykernel installed in your venv, it will complains when you run python -m ipykernel install --name=ml_skt_book under it, as the venv doesn't share packages with the base environment. So you need to install again in venv.

Did you tried to install Jupyter (pip install jupyter) in your desired venv, and then start Jupyter as usual from this venv?<br>
I usually use PyCharm to create venvs and its console for the pip-commands. If I use then Juypter inside those environments, there are only the packages according avaible, which are installed in this venv.<br><br>
I think the main issue is because you link to the jupyter notebook kernel, which sounds like one from the basic installation of Python on your system.