Unpin numpy 2 #3115
Unpin numpy 2 #3115
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #3115 +/- ##
==========================================
- Coverage 76.31% 73.98% -2.34%
==========================================
Files 109 109
Lines 12515 12438 -77
==========================================
- Hits 9551 9202 -349
- Misses 2964 3236 +272
|
| @@ -81,7 +81,7 @@ def is_constant( | |||
| def _(a: NDArray, axis: Literal[0, 1] | None = None) -> bool | NDArray[np.bool_]: | |||
| # Should eventually support nd, not now. | |||
| if axis is None: | |||
| return (a == a.flat[0]).all() | |||
| return bool((a == a.flat[0]).all()) | |||
There was a problem hiding this comment.
This makes the function return a bool instead of a np.bool_ scalar as the type hints say.
The two types behave almost identically but have a different repr with numpy 2, so this change passes the doctests.
| @@ -1,3 +1,5 @@ | |||
| """Geary's C autocorrelation.""" | |||
There was a problem hiding this comment.
I didn’t change anything in the two metrics modules, just reordered and reformatted (and replaced the removed np.float_ with np.float64, which in numpy 1.x are identical)
| >>> int(adata.obs['n_genes'].min()) | ||
| 1 |
There was a problem hiding this comment.
more repr changes: this is shown as np.int64(1) without the cast to int
| @@ -173,7 +173,7 @@ def test_download_failure(): | |||
| *DS_MARKS[ds], | |||
| ], | |||
| ) | |||
| for ds in set(sc.datasets.__all__) - DS_DYNAMIC | |||
| for ds in sorted(set(sc.datasets.__all__) - DS_DYNAMIC) | |||
There was a problem hiding this comment.
This fixes pytest -n auto. Pytest complained locally that different jobs collected “different tests” which were just differently sorted.
Sorry does this mean it doesn't work with numpy 2? |
|
It doesn’t. We could also
Sadly(?) Python doesn’t allow packages to add constraint to other packages’ dependencies, else we could tell the resolver that all currently release skmisc versions are incompatible with numpy 2 |
Are we in a rush? We are upper bounding right now anyway on the current release, no? I would just wait. |
|
We’re not upper-bounding in 1.10.1. |
|
We discussed outside @flying-sheep - we will release with the bound then and just wait here? |
|
release done, we can wait here … |
|
Looks like there is a release out for the Hopefully this fixes it |
Co-authored-by: Philipp A <[email protected]>
Other things done:
Rename some variables and reorder functions so the diff between the metrics is minimal for a future unification
Skip seurat v3 tests with numpy 2 because skmisc isn’t ready: Support for numpy 2.0 has2k1/scikit-misc#34
This will skip the seurat v3 tests for all but the minimum versions test job for now, but I think that’s OK?