初始化/导入子模块而不会污染全局名称空间

Question

Typically when I import modules, I prefer to just import the top module and treat everything as a member of that, rather than importing each function I need into the global namespace, ie:

import os
os.walk()

instead of:

from os import walk
walk()

However, I recently came across modules that have submodules that seemingly have to be imported globally in order to run. For example, the Scikit-Learn module contains the "cluster" submodule. I want to do this:

import sklearn as skl
skl.cluster.KMeans(...)

but this throws an error because I didn't initialize sklearn.cluster . Instead I have seen this:

from sklearn import cluster
cluster.KMeans(...)

I really don't like this because I don't like polluting my global namespace. Is there any way around this? I tried this:

import sklearn as skl
import skl.cluster

but that doesn't work either. How can I initialize a submodule without importing it globally?

Answer 1

import sklearn as skl
import sklearn.cluster

The name after the import has to be the actual name of the module, not an alias. This does mean that you will end up with the sklearn name as well as the skl name bound to the sklearn module object.

There is no import that will initialize sklearn.cluster and bind the skl name to the sklearn module. The closest you can do is del sklearn after importing the submodules, or shove the submodule initialization imports somewhere where they won't bind names in this namespace, like a dedicated function or module or something (but hiding imports like that has its own issues), or use importlib.import_module to load the submodules.