多个版本的 conda 和 pip 封装

Question

Python, problem with different versions of packages. It often happens that package developers specify obsolete versions of packages in their dependencies.

Is there a relatively simple (without editing the source code of the packages) way to specify that a particular package will have the old version of the dependency that it needs only, and not so that it is installed for the entire project. For conda and pip.

Now you have to either have several versions of the kernels, and restart all the necessary code in jupiterlab, or abandon some of the packages.

Example. I have a skearn.

import sklearn
sklearn.__version__

'1.1.1'

I want to install the auto-sklearn library. But in https://github.com/automl/auto-sklearn/blob/development/requirements.txt :

scikit-learn>=0.24.0,<0.25.0

This version is obsolete, and the dependency is explicitly specified, version 1.1.1 will be overwritten by 0.25.0. I want auto-sklearn to have its old sklearn dependency.

Answer 1

No, there is not a way to have one package reference one version of a dependency while all other packages use a different version.

Rather, use a separate environment to isolate different sets of packages that work together for different tasks.