Fix numpy doctest #176
Fix numpy doctest #176
Conversation
replace np.float_ with np.float64. The minimum numpy version 1.16 has both as aliases, the newer version numpy>=2.0 don't support np.float_ anymore.
[NEP51](https://numpy.org/neps/nep-0051-scalar-representation.html#backward-compatibility) changes the representation of scalars. While the long term strategy should be a comparison with the new form of representation, the discussion in [issue 24470](numpy/numpy#24470) seems to suggest to use an intermediate step while both, numpy-1 and numpy-2 are supported.
There was a problem hiding this comment.
Thanks! I'm a bit worried about always changing the print option. We really only need this when running the tests. I have some local changes where this is under the control of an environment variable:
if os.environ.get('NAVIS_TEST_ENV', '').lower() == 'true':
import numpy as np
np.set_printoptions(legacy="1.25")
This would then require also changes to test_package.yml and pytest.ini to set that environment variable. Alternatively, you could invoke the set_printoptions in conftest.py - I haven't tried that though.
navis/morpho/mmetrics.py
Outdated
| # Set up numpy number representation, see NEP51 | ||
| # Once numpy<=2 is dropped from requirements, the doctest comparissons | ||
| # should become `np.float64(1.074)` instead of `1.074` | ||
| if int(np.version.version.split('.')[0])>=2: |
There was a problem hiding this comment.
I would put this in config.py - feels more appropriate when adjusting global settings?
There was a problem hiding this comment.
I made the suggested changes.
In this scenario, the whole code basis gets switched to a different representation of np values. This means, in theory there could a piece of code that behaves differently in the NAVIS_TEST_ENV than it does whenever you run the code without that variable set to true. I can't think of a case where this would matter. Since you were worried about "global" changes to the representation / print option, I also made an alternative PR #177, where the change of number representation should only change within the pydoctest.
This PR potentially improves the numpy>=2.0 compatibility
np.float_as a dtype (and the minimum numpy version 1.16 seems to support both)np.float64(1.074), which fails the doctest when compared to1.074). While numpy<2.0 is still an accepted dependency, I change the representation back to python scalars to pass the tests.This problem came up in #175, but is a separate issue, hence the separate PR.