在运行笔记本代码之前，如何让我的 Jupyter 笔记本服务器运行任意 Python 代码？

Question

I'm trying to replicate the functionality of the code editor on a platform I was previously using called Odoo.sh. The platform would let me create a .ipynb notebook, but in the cells I could reference pre-set variables which required no boilerplate code inside of the notebook. Extremely convenient.

If you're familiar with Odoo, it was like having odoo-bin shell be implicitly run before executing any of the cells inside the notebook. It was wonderful to work with, but Odoo.sh is proprietary, so I'm trying to replicate the same functionality on my local machine.

A minimal example of what I'm going for here would be to have the following python code run before executing any of my .ipynb notebook file's cells.

example_value = False

def example_func():
    global example_value
    example_value = True

example_func()

So that inside of any notebook's cells I could simply run something like example_value and get an output of True .

In the case of Odoo.sh it almost seemed like there was a special custom kernel set up that was nothing more than a regular Python 3 kernel with some initialization code. This may be exactly what was going on, but I don't know enough about how Jupyter works to know for myself. How do I replicate this functionality?

Answer 1

I figured it out! You need to create a custom kernel, but for this use case you can just reuse the default IPython kernel and just pass some variables into the user namespace.

First, create a Python file for your kernel. Let's use test_kernel.py . Here are the contents:

from ipykernel.ipkernel import IPythonKernel 
from ipykernel.kernelapp import IPKernelApp


if __name__ == "__main__":
    example_value = False

    def example_func():
        global example_value
        example_value = True

    example_func()

    IPKernelApp.launch_instance(
        kernel_class=IPythonKernel,
        user_ns={"example_value": example_value})

See how the arbitrary code from the question is run before launching the kernel instance. Using the user_ns argument, we can pass arbitrary data to the user environment.

To get our kernel up and running we need to make a test directory and then a test/kernel.json file. It will have these contents:

{
  "argv": ["python", "-m", "test_kernel", "-f", "{connection_file}"],
  "display_name": "Test"
}

Let's install that bad boy. Run jupyter kernelspec install --user test . In that command, test is the name of the directory we created. The --user argument makes Jupyter install the kernel only for the current user. You don't have to use it if you don't want to.

Now we should be good to go! Start things up with jupyter notebook and you will see your new kernel is available to use when using notebooks. And check it out, we can see the variable we passed into the namespace:

Last of all, be sure to note that in order for this to work your test_kernel.py file will need to be somewhere where Python can import it. I'm not an expert on this, but from a bit of Googling I took this to mean that the directory containing the file should either be the current working directory or be in your PATH .