在一个函数中导​​入多个python模块，以轻松设置熟悉的环境

Question

I use a lot of Jupyter notebooks for my work to design systems. I find myself doing the same things to set up my environment for each notebook I start, soo I figured I would make a function that does all that for me. It should import a list of modules into the desired namespace. For instance, say this function exists at: my_package.setup_workspace:

def imports():
    import numpy as np

I would then like to be able to setup my workspace on a new jupyter notebook with:

import my_package
my_package.setup_workspace
np.array(["I wish this could work!"])

But when I do that, python reports that it can't find np, so the imports() function wasn't able to properly import numpy into the new jupyter workspace.

Answer 1

When you do an import in a function you import things in the local namespace of the function, not the module global one. Unless you declare that the variable are global, but then they need to already exit in outer scope. This allow you to reuse variable name across function without "contaminating" outer scope.

You have a couple ways of going forward:

First self contained way see this twitter thread and this solution, to define your own magic. Then you can then do %imports in a notebook to replace the current cell by a bunch of imports. Advantage is you can pass options, but it works only in IPython.

Second, simply remove the function and do all your setup at the top level of my_package.py , then do from my_package import * . import * are frowned upon in the Python world, but if it's just for you, you should be safe.

Third, look at what %run -i in IPython can do, it's hackish but should work, as it can execute and external scripts in the current namespace.

There is a couple of other crazy way you can investigate once you want to dive deep into how python is working: You can, for example actually manually extract setup_workspace 's __code__ object; modify it and re-exec it to not create a local scope, but that's probably not what you'd looking for. It is possible though, but may not be portable across Python versions.