Closed MalteEbner closed 1 month ago
All modified and coverable lines are covered by tests :white_check_mark:
Comparison is base (
6bd7553
) 84.38% compared to head (3f1bc45
) 84.38%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
How does https://github.com/lightly-ai/lightly/pull/1559 affect this ?
@MalteEbner what is the state of this PR?
IMO there are two options:
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.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.
This has been fixed in https://github.com/lightly-ai/lightly/pull/1625
We still support numpy<1.20 and only import from numpy.typing when type-checking.
We are using the
numpy.typing
subpackage, which is only available in numpy >= 1.20: https://numpy.org/devdocs/reference/typing.html#module-numpy.typingThus we need to update the minimum requirement.