Remove np.bool8 support (deprecated) #6633
Conversation
|
One remaining warning must be resolved, which stems from upstream in scipy, and has been fixed in scipy/scipy#17467. |
| def test_very_large_labels(): | ||
| imax = np.iinfo(np.int64).max | ||
| imax = np.iinfo(np.int32).max | ||
| labels = np.array([0, 1, imax, 42, 42], dtype=np.int64) | ||
| output, fw, inv = relabel_sequential(labels, offset=imax) | ||
| assert np.max(output) == imax + 2 |
There was a problem hiding this comment.
Maybe I'm misunderstanding what was intended by this test, but if we specify imax = np.iinfo(np.int64).max and then add an offset to the new labels of offset=imax, numpy can't accurately distinguish between imax + x where x > 0. I've changed the test to use a "smaller" large number so the np.int64 doesn't overflow immediately within relabel_sequential.
| rtol = 1e-2 if dtype == np.float32 else 6e-3 | ||
| assert_allclose(dE2, data['dE'], rtol=rtol) |
There was a problem hiding this comment.
Were these relative tolerances set empirically or some other way? The tests fail for only the first entry in dE2 with numpy v1.24.0.
There was a problem hiding this comment.
pretty sure it is empirical (I think I added the 1e-2 for float32 at some point, but not sure about the original choice of 1e-4). I don't know why this particular function seems to have relatively low precision.
|
Happy for #6637 to supersede this PR if that works! |
|
I suspect we need both; will wait for @grlee77 to chime in. |
| @@ -909,7 +909,7 @@ cdef class Cascade: | |||
| feature_number = int(internal_nodes[0]) | |||
| # list() is for Python3 fix here | |||
| lut_array = list(map(lambda x: int(x), internal_nodes[1:])) | |||
| lut = np.asarray(lut_array, dtype='uint32') | |||
| lut = np.array(lut_array).astype(np.uint32) | |||
There was a problem hiding this comment.
|
closing as this has been folded into #6637 |
Description
numpy/numpy#22607 deprecated use of the
np.bool8dtype. This PR simply removes the relevant bits of skimage that concern that dtype, so deprecation warnings don't cause downstream packages to run into deprecation warnings/errors.This was discovered in dev-dependency test logs while working on spacetelescope/jdaviz#1834.
For reviewers
later.
__init__.py.doc/release/release_dev.rst.example, to backport to v0.19.x after merging, add the following in a PR
comment:
@meeseeksdev backport to v0.19.xrun-benchmarklabel. To rerun, the labelcan be removed and then added again. The benchmark output can be checked in
the "Actions" tab.