upgrade requirements/base.txt minimum numpy version to 1.20 for typing

[MalteEbner]

We are using the numpy.typing subpackage, which is only available in numpy >= 1.20: https://numpy.org/devdocs/reference/typing.html#module-numpy.typing

Thus we need to update the minimum requirement.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (6bd7553) 84.38% compared to head (3f1bc45) 84.38%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1490   +/-   ##
=======================================
  Coverage   84.38%   84.38%           
=======================================
  Files         139      139           
  Lines        5756     5756           
=======================================
  Hits         4857     4857           
  Misses        899      899           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[IgorSusmelj]

Let's not merge this as of now. We need a proper strategy to deal with the dependencies.

For context:

• We just added https://github.com/lightly-ai/lightly/blob/master/requirements/minimal_requirements.txt which contains the actual minimum requirements needed to use all the features. We have a new test for this in the CI pipeline
• We did not want to change all the base requirements yet, as we have many users using only very specific features due to the modular way lightly works. We will have to figure out how we want to deal with that. The issue is that we accidentally broke many requirements because our previous tests were always installing latest packages. So we never tested if the requirements we once established are still valid or not.

[japrescott]

lets quickly review the minimum requirements

[SauravMaheshkar]

How does #1559 affect this ?

[guarin]

@MalteEbner what is the state of this PR?

IMO there are two options:

• Handle from numpy.typing import NDArray depending on whether it is available or not. If it is not available we could define a generic alias for NDArray instead. Not sure how that would work for older Python versions though.
• Increase minimal numpy version in pyproject.toml. Numpy 1.18 has reached end of life >2 years ago.

The current state with numpy>=1.18.1, <2 as default requirement and numpy==1.26.6 as minimal requirement is not ideal.

[guarin]

This has been fixed in #1625

We still support numpy<1.20 and only import from numpy.typing when type-checking.