Evgeni Burovski

towards gh-391

Pain points in using the test suite

8

There were reports of several pain points with using the test suite. This is a central tracker for potential improvements of the test suite ergonomics. Things previously reported: - difficulty...

Jax float64 precision issues do not play ball with hypothesis

3

A typical example is (test_diff): ``` self = , val = 2.112233982580733, val_0d = Array(2.1122339, dtype=float32) strategy = FloatStrategy(min_value=2.0, max_value=64.0, allow_nan=False, smallest_nonzero_magnitude=2.2250738585072014e-308) def check_set_value(self, val, val_0d, strategy): if val ==...

ENH: torch: allow negative indices in `take` and `take_along_axis`

3

fixes https://github.com/data-apis/array-api-compat/issues/360 A matching test update: https://github.com/data-apis/array-api-tests/pull/397

ENH: use torch.clamp for wrapped_torch.clip

1

Closes gh-350 Otherwise, the version which emulates "clip" fails with torch.vmap. ------------- This patch is less innocuous than it looks, because it changes the promotion rules. Previously, `min` and `max`...

CI broken on python 3.10

2

Recent CI runs (gh-353, gh-361) are all broken on python 3.10, with tracebacks pointing to `hypothesis` internals. E.g. https://github.com/data-apis/array-api-compat/actions/runs/19609218701/job/56152540302?pr=361 Bisecting hypothsis versions locally, it appears that hypothesis `6.144` is the...

**Describe the issue**: Indexing a complex-valued array with an integer array emits a numpy `ComplexWarning` even if the indexing result is correct. **Minimal Complete Verifiable Example**: ```python >>> import dask.array...

ENH: linalg: return complex arrays from eigvals/eigvecs

2

A (simplest) companion to https://github.com/numpy/numpy/issues/29000 : change the dtype of `eig/eigvals` returns from _maybe float if starts align, else complex_ to always complex. As an opinionated pitch from the long...

Make `eig/eigvals` always return complex eigenvalues

20

For a real-valued input, NumPy's eigenvalue routines currently may return either real or complex eigenvalues: both `eig` and `eigvals` check for imaginary parts being exactly zero, and downcast the output...

ENH: linalg/SVD: move the batching loop to C

1

#### Reference issue #### What does this implement/fix? For `linalg.svd` and `linalg.svdvals`, wrap LAPACK functions directly, instead of a python loop over the f2py wrapped routines. The C side uses...