TST, MAINT: run optimize array API tests and fix chandrupatla
#20737
Conversation
* Addresses some of my points at: scipy#20085 (review) and seems to fix about 55 GPU-based array API test failures [skip cirrus] [skip circle]
|
Matt will probably find a few more fixes that I missed, and/or slightly adjust these ones. Feel free to push over, close, etc. |
| @@ -655,11 +655,11 @@ def test_convergence(self, xp): | |||
| x1, x2 = bracket | |||
| f0 = xp_minimum(xp.abs(self.f(x1, *args)), xp.abs(self.f(x2, *args))) | |||
| res1 = _chandrupatla_root(self.f, *bracket, **kwargs) | |||
| xp_assert_less(np.abs(res1.fun), 1e-3*f0) | |||
| xp_assert_less(xp.abs(res1.fun), 1e-3*f0) | |||
There was a problem hiding this comment.
I was confused about why this was passing with torch CPU, but ufuncs seem to preserve type.
import numpy as np
import torch
x = torch.asarray([-1, 1.])
np.abs(x) # tensor([1., 1.])Same as exp, etc.
There was a problem hiding this comment.
I was confused about why this was passing with
torchCPU, but ufuncs seem to preserve type.
Only when the input array type implements __array_wrap__ or some other such method. torch.Tensor.__array_wrap__ exists, so the return value remains a torch.Tensor.
mdhaber
force-pushed
the
treddy_arr_api_gpu_fixes
branch
from
ba7f43d
to
dce9777
Compare
rgommers
changed the title
chandrupatla
rgommers
added
the
array types
There was a problem hiding this comment.
LGTM too, thanks @tylerjereddy and @mdhaber.
I found a few more issues in gh-20085 that are new with CuPy. So I'll go ahead and merge this, that makes it easier to finalize gh-20085.
[skip cirrus] [skip circle]