ENH: stats.pearsonr: add array API support #20284

mdhaber · 2024-03-19T02:27:22Z

Reference issue

gh-20137
Closes gh-20324

What does this implement/fix?

This explores the addition of Array API support to scipy.stats.pearsonr. Only the last commit is relevant; the others are from gh-20137.

Additional information

Need to resolve merge conflicts and skip a test on 32-bit, but otherwise, I think this is ready to go.

Old news:

Most of this will be pretty straightforward. There are some little things I'll want to address later (e.g. previously, pearsonr converted inputs to be at least float64, but I imagine we'd want to respect dtype with array API), but there is one big question for now:

Calculation of the p-value currently relies on the incomplete beta function, which is not among the special functions for which we have experimental array API support (gh-19023). Even if it were, calculation of the p-value currently relies on the distribution infrastructure. In any case, calculation of the statistic with an alternative array backend can easily be done now, but calculation of the p-value with an alternative array backend will take more time.

For vectorized calculations, I think there is still value in calculating the statistic with the alternative array backend, then converting the statistic (which has been reduced by at least one dimension) to a NumPy array for calculation of the p-value. Are there objections to this?

I know there have been objections to converting non-NumPy arrays to NumPy arrays for running compiled code, but I think this is a little different, since the statistic might not even be an array after the reducing operation, and if it is, it's smaller than the original array.

[skip actions] [skip cirrus]

lucascolley · 2024-03-20T12:29:33Z

For vectorized calculations, I think there is still value in calculating the statistic with the alternative array backend, then converting the statistic (which has been reduced by at least one dimension) to a NumPy array for calculation of the p-value. Are there objections to this?

This sounds right to me. As long as we avoid converting back-and-forth multiple times, one pair of conversions is necessary (unless someone decides to write the special functions in pure Python 😅 (or the special extension gets into the standard...) and the distn. infra gets support).

mdhaber · 2024-03-20T14:23:26Z

Yes, in the near term, the new distribution infrastructure will be able to evaluate the special functions in an array API compatible way, and I will give the resampling methods array API support soon, too. So this would just be temporary, probably for one release only, if that.

Further out, yeah, the special function array API extension (data-apis/array-api#725) would speed things up considerably. (Oops looks like you mentioned it in an update, but maybe good to have the link here for others.)

So in the meantime, is there a canonical way to do the conversion? np.asarray alone doesn't cut it, right?

lucascolley · 2024-03-20T15:01:31Z

So in the meantime, is there a canonical way to do the conversion? np.asarray alone doesn't cut it, right?

Just use np.asarray for now is my suggestion (and xp.asarray at the end). There is no universally portable solution, but the idiomatic way going forward will be to use from_dlpack and co. That has seen updates quite recently and we don't use it anywhere yet, so it's probably easiest to update wholesale across submodules at a later date. Plus, np.asarray is sufficient for the backends with which we test in CI.

lucascolley

looking pretty good!

scipy/stats/_morestats.py

scipy/stats/_stats_py.py

[skip ci]

scipy/stats/_stats_py.py

lucascolley

do you want to add the CI check in this PR?

scipy/stats/_stats_py.py

mdhaber · 2024-03-21T13:11:22Z

scipy/stats/tests/test_stats.py

+        res_ci = res.confidence_interval()
+        ref_ci = ref.confidence_interval()
+        xp_assert_close(res_ci.low, xp.asarray(ref_ci.low))
+        xp_assert_close(res_ci.high, xp.asarray(ref_ci.high))


Could use input validation tests. Maybe should also check edge cases like length-2 input, constant input, etc... with array API but I imagine that someday we'll just want to have the option of running all tests with non-numpy arrays, no? I don't think we should duplicate existing tests for array API, right? Should I convert all the tests now?

Yeah, better to convert existing tests where possible, perhaps splitting into 2 and running one with np_only.

e.g.

scipy/scipy/fft/tests/test_basic.py

Lines 332 to 368 in 235602c

@skip_if_array_api(np_only=True)

@pytest.mark.parametrize("dtype", [np.float16, np.longdouble])

def test_dtypes_nonstandard(self, dtype):

x = random(30).astype(dtype)

out_dtypes = {np.float16: np.complex64, np.longdouble: np.clongdouble}

x_complex = x.astype(out_dtypes[dtype])

res_fft = fft.ifft(fft.fft(x))

res_rfft = fft.irfft(fft.rfft(x))

res_hfft = fft.hfft(fft.ihfft(x), x.shape[0])

# Check both numerical results and exact dtype matches

assert_array_almost_equal(res_fft, x_complex)

assert_array_almost_equal(res_rfft, x)

assert_array_almost_equal(res_hfft, x)

assert res_fft.dtype == x_complex.dtype

assert res_rfft.dtype == np.result_type(np.float32, x.dtype)

assert res_hfft.dtype == np.result_type(np.float32, x.dtype)

@pytest.mark.parametrize("dtype", ["float32", "float64"])

def test_dtypes_real(self, dtype, xp):

x = xp.asarray(random(30), dtype=getattr(xp, dtype))

res_rfft = fft.irfft(fft.rfft(x))

res_hfft = fft.hfft(fft.ihfft(x), x.shape[0])

# Check both numerical results and exact dtype matches

rtol = {"float32": 1.2e-4, "float64": 1e-8}[dtype]

xp_assert_close(res_rfft, x, rtol=rtol, atol=0)

xp_assert_close(res_hfft, x, rtol=rtol, atol=0)

@pytest.mark.parametrize("dtype", ["complex64", "complex128"])

def test_dtypes_complex(self, dtype, xp):

x = xp.asarray(random(30), dtype=getattr(xp, dtype))

res_fft = fft.ifft(fft.fft(x))

# Check both numerical results and exact dtype matches

rtol = {"complex64": 1.2e-4, "complex128": 1e-8}[dtype]

xp_assert_close(res_fft, x, rtol=rtol, atol=0)

Should I convert all the tests now?

If you want - it should be done eventually at least

Do we still need this separate test after converting the others?

I thought about that. I'll think about it some more before my next push.

For old functions like this, without really thinking through the existing test suite from the bottom up, it's hard to be sure that the tests are sufficient. So I think it's valuable to have a property-based test like this. To be more thorough, I might try using hypothesis.

mdhaber · 2024-03-21T16:40:38Z

do you want to add the CI check in this PR?

Oops, missed that.
So maybe:

        python dev.py --no-build test -b all -t scipy.stats.tests.test_stats -- --durations 3 --timeout=60

at the end of array_api.yml?

lucascolley · 2024-03-21T16:45:27Z

Yeah, looks correct 👍

scipy/stats/tests/test_stats.py

mdhaber · 2024-03-22T01:27:52Z

I had to create my own _move_axis_to_end function to get past that failure and see what else was wrong. I also added a _clip function and fixed the uses of [()] to get all the tests passing, so I went ahead and left everything in there. We can replace _move_axis_to_end and _clip when array_api_strict is updated.

Regarding [()] - These four lines used to appear all over the place in stats to ensure that we return scalars instead of 0D arrays:

if res.ndim == 0:
    return res.item
else:
    return res

Usingres[()] is not only more compact, but it also ensures that the values are NumPy scalars instead of Python scalars. This is more consistent with NumPy's behavior.

array_api_strict allows the use of [()] on 0D arrays, but raises an IndexError when the array has 1 or more dimensions, so I replaced this idiom with a ternary like res[()] if res.ndim == 0 else res. This is not so bad, but is there another option I'm missing?

lucascolley · 2024-03-22T09:30:26Z

array_api_strict allows the use of [()] on 0D arrays, but raises an IndexError when the array has 1 or more dimensions, so I replaced this idiom with a ternary like res[()] if res.ndim == 0 else res. This is not so bad, but is there another option I'm missing?

@rgommers has this point about returning scalars come up previously?

.github/workflows/array_api.yml

rgommers · 2024-03-22T21:11:10Z

@rgommers has this point about returning scalars come up previously?

I don't think so. This pattern is pretty specific to SciPy.

Usingres[()] is not only more compact, but it also ensures that the values are NumPy scalars instead of Python scalars. This is more consistent with NumPy's behavior.

Agreed, that is better.

so I replaced this idiom with a ternary like res[()] if res.ndim == 0 else res. This is not so bad, but is there another option I'm missing?

That looks okay to me; I don't think it'll get clearer/shorter than that.

lucascolley

I know this isn't finished yet but the array API changes look good to me (I do need to check CUDA though). If a stats maintainer could give a 👍 once it is ready that would be good.

@asmeurer out of interest, do you have a rough ETA for 2023.12 support in array-api-strict?

lucascolley · 2024-03-25T20:38:57Z

scipy/stats/_stats_py.py

+
+def _move_axis_to_end(x, source, xp):
+    axes = list(range(x.ndim))
+    temp =  axes.pop(source)


Suggested change

temp = axes.pop(source)

temp = axes.pop(source)

mdhaber · 2024-03-25T21:11:36Z

Actually I do think I'd like to consider this done, unless there was something besides the array API test? After I do a few of these, I'll come back and replace the separate array API test with a more thorough test with hypothesis. It would probably be best to do those at the same time anyway, since I think we can share some things (e.g. strategys) between the tests.

mdhaber · 2024-03-25T21:38:47Z

@tupui Would you be interested in checking that I didn't change anything from a stats perspective?

lucascolley · 2024-03-26T00:03:17Z

I'm having lots of problems with trying to get CUDA working again after changing some drivers while working on JAX. I'll try to get it resolved but someone else may have to check GPU.

lucascolley · 2024-03-26T00:18:47Z

Okay, I got PyTorch CUDA working, quite a few failures, see below @mdhaber. CuPy will have to wait (cupy/cupy#8260 ...) but that's alright.

traceback

______________________________________________________________________________ TestPearsonr.test_r_almost_exactly_pos1[torch] ______________________________________________________________________________
scipy/stats/tests/test_stats.py:361: in test_r_almost_exactly_pos1
    r, prob = stats.pearsonr(a, a)
        a          = tensor([0., 1., 2.], device='cuda:0')
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd2991190>
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 0.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd4213490>
        dtype      = torch.float32
        method     = None
        n          = 3
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([1.4142], device='cuda:0')
        normym     = tensor([1.4142], device='cuda:0')
        one        = tensor(1., device='cuda:0')
        r          = tensor(1., device='cuda:0')
        threshold  = 6.4155305118844185e-06
        x          = tensor([0., 1., 2.], device='cuda:0')
        xm         = tensor([-1.,  0.,  1.], device='cuda:0')
        xmax       = tensor([1.], device='cuda:0')
        xmean      = tensor([1.], device='cuda:0')
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([0., 1., 2.], device='cuda:0')
        ym         = tensor([-1.,  0.,  1.], device='cuda:0')
        ymax       = tensor([1.], device='cuda:0')
        ymean      = tensor([1.], device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(1., device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(1., device='cuda:0'),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(1., device='cuda:0')]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(1., device='cuda:0'),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(1., device='cuda:0'),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(1., device='cuda:0')
______________________________________________________________________________ TestPearsonr.test_r_almost_exactly_neg1[torch] ______________________________________________________________________________
scipy/stats/tests/test_stats.py:370: in test_r_almost_exactly_neg1
    r, prob = stats.pearsonr(a, -a)
        a          = tensor([0., 1., 2.], device='cuda:0')
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd299d1d0>
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 0.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd3eed750>
        dtype      = torch.float32
        method     = None
        n          = 3
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([1.4142], device='cuda:0')
        normym     = tensor([1.4142], device='cuda:0')
        one        = tensor(1., device='cuda:0')
        r          = tensor(-1., device='cuda:0')
        threshold  = 6.4155305118844185e-06
        x          = tensor([0., 1., 2.], device='cuda:0')
        xm         = tensor([-1.,  0.,  1.], device='cuda:0')
        xmax       = tensor([1.], device='cuda:0')
        xmean      = tensor([1.], device='cuda:0')
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([-0., -1., -2.], device='cuda:0')
        ym         = tensor([ 1.,  0., -1.], device='cuda:0')
        ymax       = tensor([1.], device='cuda:0')
        ymean      = tensor([-1.], device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(-1., device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(-1., device='cuda:0'),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(-1., device='cuda:0')]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(-1., device='cuda:0'),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(-1., device='cuda:0'),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(-1., device='cuda:0')
______________________________________________________________________________________ TestPearsonr.test_basic[torch] ______________________________________________________________________________________
scipy/stats/tests/test_stats.py:382: in test_basic
    r, prob = stats.pearsonr(a, b)
        a          = tensor([-1,  0,  1], device='cuda:0')
        b          = tensor([0, 0, 3], device='cuda:0')
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd299e810>
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 0.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9d9f89ab90>
        dtype      = torch.float64
        method     = None
        n          = 3
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([1.4142], device='cuda:0', dtype=torch.float64)
        normym     = tensor([2.4495], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.8660, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([-1.,  0.,  1.], device='cuda:0', dtype=torch.float64)
        xm         = tensor([-1.,  0.,  1.], device='cuda:0', dtype=torch.float64)
        xmax       = tensor([1.], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([0.], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([0., 0., 3.], device='cuda:0', dtype=torch.float64)
        ym         = tensor([-1., -1.,  2.], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([2.], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([1.], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.8660, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.8660, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.8660, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.8660, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.8660, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.8660, device='cuda:0', dtype=torch.float64)
_________________________________________________________________________________ TestPearsonr.test_constant_input[torch] __________________________________________________________________________________
scipy/stats/tests/test_stats.py:393: in test_constant_input
    r, p = stats.pearsonr(x, y)
        msg        = 'An input array is constant'
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd35d7290>
        x          = tensor([0.6670, 0.6670, 0.6670], device='cuda:0')
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([0.1230, 0.4560, 0.7890], device='cuda:0')
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 0.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(True, device='cuda:0')
        const_xy   = tensor(True, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd32fcd50>
        dtype      = torch.float32
        method     = None
        msg        = 'An input array is constant; the correlation coefficient is not defined.'
        n          = 3
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([nan], device='cuda:0')
        normym     = tensor([0.4709], device='cuda:0')
        one        = tensor(1., device='cuda:0')
        r          = tensor(nan, device='cuda:0')
        threshold  = 6.4155305118844185e-06
        x          = tensor([0.6670, 0.6670, 0.6670], device='cuda:0')
        xm         = tensor([0., 0., 0.], device='cuda:0')
        xmax       = tensor([0.], device='cuda:0')
        xmean      = tensor([0.6670], device='cuda:0')
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([0.1230, 0.4560, 0.7890], device='cuda:0')
        ym         = tensor([-3.3300e-01, -2.9802e-08,  3.3300e-01], device='cuda:0')
        ymax       = tensor([0.3330], device='cuda:0')
        ymean      = tensor([0.4560], device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(nan, device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(nan, device='cuda:0'),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(nan, device='cuda:0')]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(nan, device='cuda:0'),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(nan, device='cuda:0'),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(nan, device='cuda:0')
___________________________________________________________________________ TestPearsonr.test_near_constant_input[float32-torch] ___________________________________________________________________________
scipy/stats/tests/test_stats.py:408: in test_near_constant_input
    stats.pearsonr(x, y)
        dtype      = torch.float32
        msg        = 'An input array is nearly constant; the computed'
        npdtype    = <class 'numpy.float32'>
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd35d6fd0>
        x          = tensor([2.0000, 2.0000, 2.0000], device='cuda:0')
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([3.0000, 3.0000, 3.0000], device='cuda:0')
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 0.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd283b0d0>
        dtype      = torch.float32
        method     = None
        msg        = 'An input array is nearly constant; the computed correlation coefficient may be inaccurate.'
        n          = 3
        nconst_x   = tensor(True, device='cuda:0')
        nconst_xy  = tensor(True, device='cuda:0')
        nconst_y   = tensor(True, device='cuda:0')
        normxm     = tensor([2.3842e-07], device='cuda:0')
        normym     = tensor([1.2389e-06], device='cuda:0')
        one        = tensor(1., device='cuda:0')
        r          = tensor(0.5774, device='cuda:0')
        threshold  = 6.4155305118844185e-06
        x          = tensor([2.0000, 2.0000, 2.0000], device='cuda:0')
        xm         = tensor([0.0000e+00, 0.0000e+00, 2.3842e-07], device='cuda:0')
        xmax       = tensor([2.3842e-07], device='cuda:0')
        xmean      = tensor([2.], device='cuda:0')
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([3.0000, 3.0000, 3.0000], device='cuda:0')
        ym         = tensor([-7.1526e-07, -7.1526e-07,  7.1526e-07], device='cuda:0')
        ymax       = tensor([7.1526e-07], device='cuda:0')
        ymean      = tensor([3.0000], device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.5774, device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.5774, device='cuda:0'),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.5774, device='cuda:0')]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.5774, device='cuda:0'),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.5774, device='cuda:0'),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.5774, device='cuda:0')
___________________________________________________________________________ TestPearsonr.test_near_constant_input[float64-torch] ___________________________________________________________________________
scipy/stats/tests/test_stats.py:408: in test_near_constant_input
    stats.pearsonr(x, y)
        dtype      = torch.float64
        msg        = 'An input array is nearly constant; the computed'
        npdtype    = <class 'numpy.float64'>
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd35d7c90>
        x          = tensor([2.0000, 2.0000, 2.0000], device='cuda:0', dtype=torch.float64)
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([3.0000, 3.0000, 3.0000], device='cuda:0', dtype=torch.float64)
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 0.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd27dba90>
        dtype      = torch.float64
        method     = None
        msg        = 'An input array is nearly constant; the computed correlation coefficient may be inaccurate.'
        n          = 3
        nconst_x   = tensor(True, device='cuda:0')
        nconst_xy  = tensor(True, device='cuda:0')
        nconst_y   = tensor(True, device='cuda:0')
        normxm     = tensor([4.4409e-16], device='cuda:0', dtype=torch.float64)
        normym     = tensor([2.1756e-15], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.8165, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([2.0000, 2.0000, 2.0000], device='cuda:0', dtype=torch.float64)
        xm         = tensor([0.0000e+00, 0.0000e+00, 4.4409e-16], device='cuda:0',
       dtype=torch.float64)
        xmax       = tensor([4.4409e-16], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([2.], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([3.0000, 3.0000, 3.0000], device='cuda:0', dtype=torch.float64)
        ym         = tensor([-8.8818e-16, -8.8818e-16,  1.7764e-15], device='cuda:0',
       dtype=torch.float64)
        ymax       = tensor([1.7764e-15], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([3.0000], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.8165, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.8165, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.8165, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.8165, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.8165, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.8165, device='cuda:0', dtype=torch.float64)
_____________________________________________________________________________ TestPearsonr.test_very_small_input_values[torch] _____________________________________________________________________________
scipy/stats/tests/test_stats.py:418: in test_very_small_input_values
    r, p = stats.pearsonr(x, y)
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd35d7ad0>
        x          = tensor([0.0044, 0.0048, 0.0039, 0.0038, 0.0034], device='cuda:0',
       dtype=torch.float64)
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([2.4800e-188, 7.4100e-181, 4.0900e-208, 2.0800e-223, 2.6600e-245],
       device='cuda:0', dtype=torch.float64)
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 1.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd2f72a50>
        dtype      = torch.float64
        method     = None
        n          = 5
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([0.0011], device='cuda:0', dtype=torch.float64)
        normym     = tensor([6.6277e-181], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.7273, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([0.0044, 0.0048, 0.0039, 0.0038, 0.0034], device='cuda:0',
       dtype=torch.float64)
        xm         = tensor([ 0.0004,  0.0007, -0.0002, -0.0003, -0.0007], device='cuda:0',
       dtype=torch.float64)
        xmax       = tensor([0.0007], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([0.0041], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([2.4800e-188, 7.4100e-181, 4.0900e-208, 2.0800e-223, 2.6600e-245],
       device='cuda:0', dtype=torch.float64)
        ym         = tensor([-1.4820e-181,  5.9280e-181, -1.4820e-181, -1.4820e-181, -1.4820e-181],
       device='cuda:0', dtype=torch.float64)
        ymax       = tensor([5.9280e-181], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([1.4820e-181], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.7273, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.7273, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.7273, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.7273, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.7273, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.7273, device='cuda:0', dtype=torch.float64)
_____________________________________________________________________________ TestPearsonr.test_very_large_input_values[torch] _____________________________________________________________________________
scipy/stats/tests/test_stats.py:432: in test_very_large_input_values
    r, p = stats.pearsonr(x, y)
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd35d5090>
        x          = tensor([0.0000e+00, 0.0000e+00, 0.0000e+00, 1.0000e+90, 1.0000e+90, 1.0000e+90,
        1.0000e+90], device='cuda:0', dtype=torch.float64)
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([0.0000e+00, 1.0000e+90, 2.0000e+90, 3.0000e+90, 4.0000e+90, 5.0000e+90,
        6.0000e+90], device='cuda:0', dtype=torch.float64)
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 2.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd32fdc50>
        dtype      = torch.float64
        method     = None
        n          = 7
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([1.3093e+90], device='cuda:0', dtype=torch.float64)
        normym     = tensor([5.2915e+90], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.8660, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([0.0000e+00, 0.0000e+00, 0.0000e+00, 1.0000e+90, 1.0000e+90, 1.0000e+90,
        1.0000e+90], device='cuda:0', dtype=torch.float64)
        xm         = tensor([-5.7143e+89, -5.7143e+89, -5.7143e+89,  4.2857e+89,  4.2857e+89,
         4.2857e+89,  4.2857e+89], device='cuda:0', dtype=torch.float64)
        xmax       = tensor([5.7143e+89], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([5.7143e+89], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([0.0000e+00, 1.0000e+90, 2.0000e+90, 3.0000e+90, 4.0000e+90, 5.0000e+90,
        6.0000e+90], device='cuda:0', dtype=torch.float64)
        ym         = tensor([-3.0000e+90, -2.0000e+90, -1.0000e+90,  4.5231e+74,  1.0000e+90,
         2.0000e+90,  3.0000e+90], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([3.0000e+90], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([3.0000e+90], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.8660, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.8660, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.8660, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.8660, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.8660, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.8660, device='cuda:0', dtype=torch.float64)
__________________________________________________________________________ TestPearsonr.test_extremely_large_input_values[torch] ___________________________________________________________________________
scipy/stats/tests/test_stats.py:445: in test_extremely_large_input_values
    r, p = stats.pearsonr(x, y)
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd293dc50>
        x          = tensor([2.3000e+200, 4.5000e+200, 6.7000e+200, 8.0000e+200], device='cuda:0',
       dtype=torch.float64)
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([1.2000e+199, 5.5000e+200, 3.3000e+201, 1.0000e+200], device='cuda:0',
       dtype=torch.float64)
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 1.0
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd39a4750>
        dtype      = torch.float64
        method     = None
        n          = 4
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([4.3437e+200], device='cuda:0', dtype=torch.float64)
        normym     = tensor([2.6978e+201], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.3513, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([2.3000e+200, 4.5000e+200, 6.7000e+200, 8.0000e+200], device='cuda:0',
       dtype=torch.float64)
        xm         = tensor([-3.0750e+200, -8.7500e+199,  1.3250e+200,  2.6250e+200],
       device='cuda:0', dtype=torch.float64)
        xmax       = tensor([3.0750e+200], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([5.3750e+200], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([1.2000e+199, 5.5000e+200, 3.3000e+201, 1.0000e+200], device='cuda:0',
       dtype=torch.float64)
        ym         = tensor([-9.7850e+200, -4.4050e+200,  2.3095e+201, -8.9050e+200],
       device='cuda:0', dtype=torch.float64)
        ymax       = tensor([2.3095e+201], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([9.9050e+200], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.3513, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.3513, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.3513, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.3513, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.3513, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.3513, device='cuda:0', dtype=torch.float64)
_______________________________________________________________________ TestPearsonr.test_negative_correlation_pvalue_gh17795[torch] _______________________________________________________________________
scipy/stats/tests/test_stats.py:507: in test_negative_correlation_pvalue_gh17795
    test_greater = stats.pearsonr(x, y, alternative='greater')
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd292c590>
        x          = tensor([0., 1., 2., 3., 4., 5., 6., 7., 8., 9.], device='cuda:0')
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([-0., -1., -2., -3., -4., -5., -6., -7., -8., -9.], device='cuda:0')
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 4.0
        alternative = 'greater'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9d9faa9250>
        dtype      = torch.float32
        method     = None
        n          = 10
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([9.0830], device='cuda:0')
        normym     = tensor([9.0830], device='cuda:0')
        one        = tensor(1., device='cuda:0')
        r          = tensor(-1., device='cuda:0')
        threshold  = 6.4155305118844185e-06
        x          = tensor([0., 1., 2., 3., 4., 5., 6., 7., 8., 9.], device='cuda:0')
        xm         = tensor([-4.5000, -3.5000, -2.5000, -1.5000, -0.5000,  0.5000,  1.5000,  2.5000,
         3.5000,  4.5000], device='cuda:0')
        xmax       = tensor([4.5000], device='cuda:0')
        xmean      = tensor([4.5000], device='cuda:0')
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([-0., -1., -2., -3., -4., -5., -6., -7., -8., -9.], device='cuda:0')
        ym         = tensor([ 4.5000,  3.5000,  2.5000,  1.5000,  0.5000, -0.5000, -1.5000, -2.5000,
        -3.5000, -4.5000], device='cuda:0')
        ymax       = tensor([4.5000], device='cuda:0')
        ymean      = tensor([-4.5000], device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(-1., device='cuda:0')
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(-1., device='cuda:0'),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(-1., device='cuda:0')]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(-1., device='cuda:0'),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(-1., device='cuda:0'),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(-1., device='cuda:0')
_________________________________________________________________________ TestPearsonr.test_length3_r_exactly_negative_one[torch] __________________________________________________________________________
scipy/stats/tests/test_stats.py:515: in test_length3_r_exactly_negative_one
    res = stats.pearsonr(x, y)
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd292d790>
        x          = tensor([1, 2, 3], device='cuda:0')
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = tensor([  5,  -4, -13], device='cuda:0')
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 0.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd2da0e10>
        dtype      = torch.float64
        method     = None
        n          = 3
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([1.4142], device='cuda:0', dtype=torch.float64)
        normym     = tensor([12.7279], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(-1.0000, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([1., 2., 3.], device='cuda:0', dtype=torch.float64)
        xm         = tensor([-1.,  0.,  1.], device='cuda:0', dtype=torch.float64)
        xmax       = tensor([1.], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([2.], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([  5.,  -4., -13.], device='cuda:0', dtype=torch.float64)
        ym         = tensor([ 9.,  0., -9.], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([9.], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([-4.], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(-1.0000, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(-1.0000, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(-1.0000, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(-1.0000, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(-1.0000, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(-1.0000, device='cuda:0', dtype=torch.float64)
________________________________________________________________________________ TestPearsonr.test_nd_special_cases[torch] _________________________________________________________________________________
scipy/stats/tests/test_stats.py:649: in test_nd_special_cases
    res = stats.pearsonr(x, y0, axis=1)
        message    = 'An input array is constant'
        rng        = Generator(PCG64) at 0x7A9DD3674900
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd290bb10>
        x          = tensor([[1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [0.8902, 0.6780, 0.2770, 0.2558, 0.0457],
        [0.7438, 0.3129, 0.3581, 0.2467, 0.0846]], device='cuda:0',
       dtype=torch.float64)
        x0         = tensor([[0.5498, 0.4304, 0.5973, 0.0502, 0.9180],
        [0.8902, 0.6780, 0.2770, 0.2558, 0.0457],
        [0.7438, 0.3129, 0.3581, 0.2467, 0.0846]], device='cuda:0',
       dtype=torch.float64)
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y0         = tensor([[0.4358, 0.5452, 0.8832, 0.0956, 0.3132],
        [0.9488, 0.9578, 0.5985, 0.7457, 0.0528],
        [0.5912, 0.9622, 0.0335, 0.1115, 0.6736]], device='cuda:0',
       dtype=torch.float64)
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 1.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 1
        const_x    = tensor([ True, False, False], device='cuda:0')
        const_xy   = tensor([ True, False, False], device='cuda:0')
        const_y    = tensor([False, False, False], device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9d9d00b010>
        dtype      = torch.float64
        method     = None
        msg        = 'An input array is constant; the correlation coefficient is not defined.'
        n          = 5
        nconst_x   = tensor([False, False, False], device='cuda:0')
        nconst_xy  = tensor([False, False, False], device='cuda:0')
        nconst_y   = tensor([False, False, False], device='cuda:0')
        normxm     = tensor([[   nan],
        [0.6891],
        [0.4874]], device='cuda:0', dtype=torch.float64)
        normym     = tensor([[0.5841],
        [0.7429],
        [0.7857]], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor([   nan, 0.8490, 0.0233], device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([[1.0000, 1.0000, 1.0000, 1.0000, 1.0000],
        [0.8902, 0.6780, 0.2770, 0.2558, 0.0457],
        [0.7438, 0.3129, 0.3581, 0.2467, 0.0846]], device='cuda:0',
       dtype=torch.float64)
        xm         = tensor([[ 0.0000,  0.0000,  0.0000,  0.0000,  0.0000],
        [ 0.4609,  0.2487, -0.1523, -0.1736, -0.3837],
        [ 0.3945, -0.0363,  0.0089, -0.1025, -0.2646]], device='cuda:0',
       dtype=torch.float64)
        xmax       = tensor([[0.0000],
        [0.4609],
        [0.3945]], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([[1.0000],
        [0.4293],
        [0.3492]], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([[0.4358, 0.5452, 0.8832, 0.0956, 0.3132],
        [0.9488, 0.9578, 0.5985, 0.7457, 0.0528],
        [0.5912, 0.9622, 0.0335, 0.1115, 0.6736]], device='cuda:0',
       dtype=torch.float64)
        ym         = tensor([[-0.0188,  0.0906,  0.4286, -0.3590, -0.1414],
        [ 0.2881,  0.2971, -0.0623,  0.0850, -0.6079],
        [ 0.1168,  0.4878, -0.4409, -0.3629,  0.1992]], device='cuda:0',
       dtype=torch.float64)
        ymax       = tensor([[0.4286],
        [0.6079],
        [0.4878]], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([[0.4546],
        [0.6607],
        [0.4744]], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor([   nan, 0.8490, 0.0233], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor([   nan, 0.8490, 0.0233], device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor([   nan, 0.8490, 0.0233], device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor([   nan, 0.8490, 0.0233], device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor([   nan, 0.8490, 0.0233], device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor([   nan, 0.8490, 0.0233], device='cuda:0', dtype=torch.float64)
________________________________________________________________________________ TestPearsonr.test_array_api[less-0-torch] _________________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'less'
        axis       = 0
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd2917310>
        x          = array([[ 1.25284449, -1.24614903, -0.53612628,  1.26534999,  0.75664989,
         0.87453957, -0.63706588,  0.30498252....7931132 ,  0.03020483,
         1.39215557,  0.11766829,  1.95888549, -0.4528613 , -0.86131051,
        -2.26395995]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[ 1.6628851 , -0.96770994,  0.26301674,  1.11258823, -0.41086372,
        -0.01066277, -1.62733296,  1.15239179....96421457,  0.91451476,
         0.3435411 ,  0.29856825, -0.1628566 ,  1.88448012,  0.90284418,
        -1.18395045]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 4.0
        alternative = 'less'
        axis       = -1
        axis_int   = 0
        const_x    = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        const_xy   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        const_y    = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd3b0a890>
        dtype      = torch.float64
        method     = None
        n          = 10
        nconst_x   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        nconst_xy  = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        nconst_y   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        normxm     = tensor([[4.0490],
        [3.4271],
        [1.5656],
        [2.5413],
        [3.1563],
        [3.9953],
        [2.6270],
        [2.3219],
        [2.6930],
        [2.3296],
        [3.7420]], device='cuda:0', dtype=torch.float64)
        normym     = tensor([[3.4730],
        [3.1413],
        [2.6578],
        [2.9953],
        [3.3712],
        [1.9682],
        [3.4098],
        [3.3884],
        [2.2028],
        [2.6459],
        [3.2847]], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor([-0.0584,  0.1020, -0.1269, -0.0048, -0.8419, -0.3980, -0.2122,  0.1346,
        -0.0625, -0.5480,  0.0699], device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([[ 1.2528, -1.1193,  2.7997, -0.0138, -0.5785,  2.5810, -0.3692,  1.1962,
         -0.0295, -0.3094],
        [...1.7200, -0.1053,  0.9599,  0.8931, -1.8272,  0.0847,
          0.0590, -2.2640]], device='cuda:0', dtype=torch.float64)
        xm         = tensor([[ 7.1185e-01, -1.6603e+00,  2.2587e+00, -5.5479e-01, -1.1195e+00,
          2.0400e+00, -9.1022e-01,  6.5516e-...         1.6217e+00, -1.0986e+00,  8.1326e-01,  7.8761e-01, -1.5354e+00]],
       device='cuda:0', dtype=torch.float64)
        xmax       = tensor([[2.2587],
        [2.3349],
        [1.0459],
        [1.3516],
        [2.0339],
        [2.1153],
        [2.2538],
        [1.7482],
        [1.5295],
        [1.1798],
        [1.6885]], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([[ 0.5410],
        [-0.1973],
        [-0.3003],
        [ 0.1769],
        [ 0.2040],
        [-0.3847],
    ...43],
        [ 0.2107],
        [-0.0961],
        [-0.0518],
        [-0.7286]], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([[ 1.6629,  1.3359, -0.6359, -0.2946, -0.2504,  0.9862,  1.8216, -0.9064,
         -0.4087, -1.4262],
        [...0.7145, -2.9010,  0.4268, -0.2500, -0.6780,  0.0285,
          0.4470, -1.1840]], device='cuda:0', dtype=torch.float64)
        ym         = tensor([[ 1.4744,  1.1475, -0.8243, -0.4830, -0.4388,  0.7978,  1.6331, -1.0949,
         -0.5971, -1.6147],
        [...0.3590, -2.5455,  0.7823,  0.1055, -0.3225,  0.3840,
          0.8025, -0.8284]], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([[1.6331],
        [1.5525],
        [1.5650],
        [1.7256],
        [1.6695],
        [1.4283],
        [2.1999],
        [2.6240],
        [1.6404],
        [1.5041],
        [2.5455]], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([[ 0.1884],
        [ 0.3718],
        [-0.2587],
        [ 0.5371],
        [ 0.3715],
        [ 0.4505],
    ...82],
        [ 0.0106],
        [ 0.2441],
        [-0.2211],
        [-0.3555]], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor([-0.0584,  0.1020, -0.1269, -0.0048, -0.8419, -0.3980, -0.2122,  0.1346,
        -0.0625, -0.5480,  0.0699], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor([-0.0584,  0.1020, -0.1269, -0.0048, -0.8419, -0.3980, -0.2122,  0.1346,
        -0.0625, -0.5480,  0.0699], device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor([-0.0584,  0.1020, -0.1269, -0.0048, -0.8419, -0.3980, -0.2122,  0.1346,
        -0.0625, -0.5480,  0.0699], device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor([-0.0584,  0.1020, -0.1269, -0.0048, -0.8419, -0.3980, -0.2122,  0.1346,
        -0.0625, -0.5480,  0.0699], device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor([-0.0584,  0.1020, -0.1269, -0.0048, -0.8419, -0.3980, -0.2122,  0.1346,
        -0.0625, -0.5480,  0.0699], device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor([-0.0584,  0.1020, -0.1269, -0.0048, -0.8419, -0.3980, -0.2122,  0.1346,
        -0.0625, -0.5480,  0.0699], device='cuda:0', dtype=torch.float64)
________________________________________________________________________________ TestPearsonr.test_array_api[less-1-torch] _________________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'less'
        axis       = 1
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd28ff490>
        x          = array([[-0.02220825,  0.58438716,  0.46411743,  0.48648984, -1.77563158,
        -1.73566365,  1.50903584, -0.74694314....42117932, -0.75821498,
         0.48163524, -0.25109435, -1.71516097, -1.21558203,  0.22234109,
         1.03320844]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[-2.19450993e-02,  2.77240701e-01,  5.03783292e-01,
        -9.47262827e-01,  9.13756817e-01, -1.54470730e+00,
...-4.28800693e-02,
        -1.40325302e+00,  1.60334059e+00,  1.73293130e+00,
        -1.58473452e+00,  1.38707952e+00]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 4.5
        alternative = 'less'
        axis       = -1
        axis_int   = 1
        const_x    = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        const_xy   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        const_y    = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9d9ca21850>
        dtype      = torch.float64
        method     = None
        n          = 11
        nconst_x   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        nconst_xy  = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        nconst_y   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        normxm     = tensor([[3.2052],
        [2.9910],
        [3.5046],
        [3.5360],
        [3.9307],
        [2.7232],
        [3.6547],
        [3.2822],
        [3.0801],
        [3.0865]], device='cuda:0', dtype=torch.float64)
        normym     = tensor([[2.5882],
        [2.5952],
        [3.7270],
        [3.2071],
        [2.9505],
        [2.6596],
        [4.1944],
        [3.9804],
        [3.1426],
        [4.0637]], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor([-0.0851, -0.1722,  0.1804,  0.0506, -0.0238, -0.5948, -0.3078, -0.1646,
         0.4597, -0.4445], device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([[-0.0222,  0.5844,  0.4641,  0.4865, -1.7756, -1.7357,  1.5090, -0.7469,
         -0.5218,  0.4139, -0.6218],
...0.4212, -0.7582,  0.4816, -0.2511, -1.7152,
         -1.2156,  0.2223,  1.0332]], device='cuda:0', dtype=torch.float64)
        xm         = tensor([[ 0.1565,  0.7631,  0.6428,  0.6652, -1.5969, -1.5569,  1.6878, -0.5682,
         -0.3430,  0.5926, -0.4430],
...0.6778, -0.5015,  0.7383,  0.0056, -1.4585,
         -0.9589,  0.4790,  1.2899]], device='cuda:0', dtype=torch.float64)
        xmax       = tensor([[1.6878],
        [1.6876],
        [1.7699],
        [1.8252],
        [2.7981],
        [1.3628],
        [1.9272],
        [1.7170],
        [1.6777],
        [1.5743]], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([[-0.1787],
        [-0.0537],
        [-0.1857],
        [-0.0418],
        [-0.4114],
        [ 0.1497],
        [-0.6473],
        [ 0.3700],
        [-0.3124],
        [-0.2567]], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([[-2.1945e-02,  2.7724e-01,  5.0378e-01, -9.4726e-01,  9.1376e-01,
         -1.5447e+00, -5.5309e-01,  1.2173e+...e-02, -1.4033e+00,  1.6033e+00,  1.7329e+00, -1.5847e+00,
          1.3871e+00]], device='cuda:0', dtype=torch.float64)
        ym         = tensor([[ 1.1971e-01,  4.1890e-01,  6.4544e-01, -8.0561e-01,  1.0554e+00,
         -1.4030e+00, -4.1143e-01,  1.3589e+...e-01, -1.5703e+00,  1.4363e+00,  1.5659e+00, -1.7517e+00,
          1.2201e+00]], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([[1.4030],
        [1.4440],
        [2.2616],
        [2.2861],
        [1.4864],
        [1.5811],
        [2.3150],
        [2.0423],
        [1.8049],
        [1.7517]], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([[-0.1417],
        [-0.2291],
        [-0.3098],
        [-0.2128],
        [ 0.1740],
        [-0.1308],
        [ 0.1177],
        [-0.3385],
        [-0.0026],
        [ 0.1670]], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor([-0.0851, -0.1722,  0.1804,  0.0506, -0.0238, -0.5948, -0.3078, -0.1646,
         0.4597, -0.4445], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor([-0.0851, -0.1722,  0.1804,  0.0506, -0.0238, -0.5948, -0.3078, -0.1646,
         0.4597, -0.4445], device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor([-0.0851, -0.1722,  0.1804,  0.0506, -0.0238, -0.5948, -0.3078, -0.1646,
         0.4597, -0.4445], device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor([-0.0851, -0.1722,  0.1804,  0.0506, -0.0238, -0.5948, -0.3078, -0.1646,
         0.4597, -0.4445], device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor([-0.0851, -0.1722,  0.1804,  0.0506, -0.0238, -0.5948, -0.3078, -0.1646,
         0.4597, -0.4445], device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor([-0.0851, -0.1722,  0.1804,  0.0506, -0.0238, -0.5948, -0.3078, -0.1646,
         0.4597, -0.4445], device='cuda:0', dtype=torch.float64)
_______________________________________________________________________________ TestPearsonr.test_array_api[less-None-torch] _______________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'less'
        axis       = None
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd28ffd90>
        x          = array([[-0.44824269,  0.97436379, -0.42351573, -2.94171864, -0.58202323,
         0.28085284,  0.45430781,  0.31106377....49004227,  0.26625433,
        -0.96454795,  1.0258759 , -0.43427223, -1.34128379, -0.47015297,
         0.64788595]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[-9.70088660e-01,  2.88587687e-01, -1.00843070e+00,
        -2.48966632e+00,  4.59649097e-02, -6.93758261e-01,
... 3.15994578e-01,
         6.59154418e-02, -5.85216863e-01,  8.33440131e-01,
         2.59484496e+00, -1.37696849e-01]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 54.0
        alternative = 'less'
        axis       = -1
        axis_int   = -1
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd2824910>
        dtype      = torch.float64
        method     = None
        n          = 110
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([10.8875], device='cuda:0', dtype=torch.float64)
        normym     = tensor([10.1264], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.0808, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([-0.4482,  0.9744, -0.4235, -2.9417, -0.5820,  0.2809,  0.4543,  0.3111,
         0.1421, -1.7510, -0.1047,  1....0,  0.2663,
        -0.9645,  1.0259, -0.4343, -1.3413, -0.4702,  0.6479], device='cuda:0',
       dtype=torch.float64)
        xm         = tensor([-0.4445,  0.9781, -0.4198, -2.9380, -0.5783,  0.2846,  0.4581,  0.3148,
         0.1458, -1.7472, -0.1009,  1....3,  0.2700,
        -0.9608,  1.0296, -0.4305, -1.3375, -0.4664,  0.6516], device='cuda:0',
       dtype=torch.float64)
        xmax       = tensor([2.9380], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([-0.0037], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([-9.7009e-01,  2.8859e-01, -1.0084e+00, -2.4897e+00,  4.5965e-02,
        -6.9376e-01, -6.9495e-01,  3.1959e-01...
         6.5915e-02, -5.8522e-01,  8.3344e-01,  2.5948e+00, -1.3770e-01],
       device='cuda:0', dtype=torch.float64)
        ym         = tensor([-0.7772,  0.4815, -0.8155, -2.2968,  0.2389, -0.5008, -0.5020,  0.5125,
         0.9768, -1.1229, -0.1276, -0....2, -1.5139,
         0.5089,  0.2588, -0.3923,  1.0263,  2.7878,  0.0552], device='cuda:0',
       dtype=torch.float64)
        ymax       = tensor([2.7878], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([-0.1929], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.0808, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.0808, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.0808, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.0808, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.0808, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.0808, device='cuda:0', dtype=torch.float64)
_______________________________________________________________________________ TestPearsonr.test_array_api[greater-0-torch] _______________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'greater'
        axis       = 0
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd28fcc50>
        x          = array([[ 0.58280528,  1.40732762,  1.83771467, -0.92741555, -0.55993253,
        -0.94497136,  0.98335686,  0.77242536....25110639,  1.26526829,
         0.56374001, -0.78703184,  0.57498194,  1.57060302, -0.23740502,
         1.19575261]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[-1.03677722,  2.11278462, -1.21363056, -0.40898022,  1.371498  ,
         1.43571859,  0.07539048,  0.16369815....44539945, -0.57875633,
        -0.0538199 ,  0.96278244,  0.38174599,  1.39144508,  0.66212325,
        -0.47453695]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 4.0
        alternative = 'greater'
        axis       = -1
        axis_int   = 0
        const_x    = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        const_xy   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        const_y    = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd3eeea50>
        dtype      = torch.float64
        method     = None
        n          = 10
        nconst_x   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        nconst_xy  = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        nconst_y   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        normxm     = tensor([[3.0652],
        [2.2787],
        [2.2938],
        [3.5812],
        [2.6015],
        [3.1080],
        [2.8945],
        [3.5220],
        [3.4725],
        [2.0002],
        [3.3038]], device='cuda:0', dtype=torch.float64)
        normym     = tensor([[3.3216],
        [3.8948],
        [2.9030],
        [1.7474],
        [2.5263],
        [2.6399],
        [3.1116],
        [2.5216],
        [3.1296],
        [2.6459],
        [2.3065]], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor([-0.2860,  0.0520, -0.0339, -0.3349, -0.0734,  0.1423,  0.5473,  0.2595,
         0.4972,  0.1216, -0.1848], device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([[ 0.5828, -0.2959, -0.7518,  0.0750,  1.7316, -0.5242, -1.6913, -1.2775,
         -0.7687,  0.7232],
        [...0.6099,  0.5591, -1.1895,  1.0491,  1.9093, -0.6670,
         -0.9693,  1.1958]], device='cuda:0', dtype=torch.float64)
        xm         = tensor([[ 0.8025, -0.0763, -0.5321,  0.2947,  1.9513, -0.3045, -1.4716, -1.0579,
         -0.5490,  0.9429],
        [...0.3596,  0.3088, -1.4398,  0.7988,  1.6590, -0.9173,
         -1.2196,  0.9455]], device='cuda:0', dtype=torch.float64)
        xmax       = tensor([[1.9513],
        [1.1946],
        [1.3628],
        [2.0397],
        [1.4952],
        [1.5523],
        [1.3597],
        [1.5237],
        [1.8142],
        [1.2867],
        [1.6590]], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([[-0.2197],
        [ 0.4198],
        [ 0.4749],
        [ 0.4020],
        [-0.2300],
        [ 0.3645],
    ...66],
        [-0.2392],
        [ 0.1903],
        [ 0.2136],
        [ 0.2503]], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([[-1.0368, -0.9876,  1.8760, -0.9882, -1.3740,  1.2065, -0.1930, -1.2058,
         -0.9408,  0.1133],
        [...0.1863,  0.9507, -1.3370,  0.3739, -0.8247,  0.1940,
          0.5080, -0.4745]], device='cuda:0', dtype=torch.float64)
        ym         = tensor([[-0.6837, -0.6346,  2.2291, -0.6352, -1.0210,  1.5596,  0.1601, -0.8528,
         -0.5878,  0.4663],
        [...0.0175,  1.1195, -1.1682,  0.5428, -0.6559,  0.3628,
          0.6768, -0.3057]], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([[2.2291],
        [2.2011],
        [1.8484],
        [1.3435],
        [1.5583],
        [1.5410],
        [1.8077],
        [1.8425],
        [2.0300],
        [1.9792],
        [1.1682]], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([[-0.3530],
        [ 0.2249],
        [ 0.1555],
        [-0.1019],
        [-0.1868],
        [ 0.0127],
    ...28],
        [ 0.3176],
        [-0.1764],
        [-0.4173],
        [-0.1688]], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor([-0.2860,  0.0520, -0.0339, -0.3349, -0.0734,  0.1423,  0.5473,  0.2595,
         0.4972,  0.1216, -0.1848], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor([-0.2860,  0.0520, -0.0339, -0.3349, -0.0734,  0.1423,  0.5473,  0.2595,
         0.4972,  0.1216, -0.1848], device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor([-0.2860,  0.0520, -0.0339, -0.3349, -0.0734,  0.1423,  0.5473,  0.2595,
         0.4972,  0.1216, -0.1848], device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor([-0.2860,  0.0520, -0.0339, -0.3349, -0.0734,  0.1423,  0.5473,  0.2595,
         0.4972,  0.1216, -0.1848], device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor([-0.2860,  0.0520, -0.0339, -0.3349, -0.0734,  0.1423,  0.5473,  0.2595,
         0.4972,  0.1216, -0.1848], device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor([-0.2860,  0.0520, -0.0339, -0.3349, -0.0734,  0.1423,  0.5473,  0.2595,
         0.4972,  0.1216, -0.1848], device='cuda:0', dtype=torch.float64)
_______________________________________________________________________________ TestPearsonr.test_array_api[greater-1-torch] _______________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'greater'
        axis       = 1
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd28fda50>
        x          = array([[-6.33197601e-01,  8.46999061e-01,  6.59769481e-01,
         7.93870740e-01,  1.22089810e+00,  1.91213040e-01,
... 1.24912577e+00,
        -7.58850672e-01,  1.12202338e+00,  2.17771622e+00,
         1.13986898e-01, -1.26385967e+00]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[-0.62174662,  0.0971421 , -0.22749994,  0.82154066, -0.16140328,
         0.37289666, -0.91485382, -0.92668902....21140984, -0.02924982,
        -1.14998464,  0.19884276, -0.17506334,  0.11813101, -1.26135631,
         0.78346058]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 4.5
        alternative = 'greater'
        axis       = -1
        axis_int   = 1
        const_x    = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        const_xy   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        const_y    = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9d9cf35810>
        dtype      = torch.float64
        method     = None
        n          = 11
        nconst_x   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        nconst_xy  = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        nconst_y   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        normxm     = tensor([[2.3134],
        [3.3703],
        [3.0174],
        [3.5588],
        [2.7893],
        [4.7450],
        [4.1314],
        [3.7639],
        [3.7632],
        [3.2933]], device='cuda:0', dtype=torch.float64)
        normym     = tensor([[2.7549],
        [2.9301],
        [3.5768],
        [2.8263],
        [2.5557],
        [4.3706],
        [2.5169],
        [2.6224],
        [2.7146],
        [2.5738]], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor([ 0.5595, -0.3820, -0.0221, -0.5843,  0.0358, -0.1136,  0.4268, -0.0320,
         0.1027, -0.2719], device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([[-6.3320e-01,  8.4700e-01,  6.5977e-01,  7.9387e-01,  1.2209e+00,
          1.9121e-01, -7.5429e-01,  2.7702e-...e+00, -7.5885e-01,  1.1220e+00,  2.1777e+00,  1.1399e-01,
         -1.2639e+00]], device='cuda:0', dtype=torch.float64)
        xm         = tensor([[-0.9913,  0.4889,  0.3016,  0.4357,  0.8628, -0.1669, -1.1124, -0.0811,
         -0.3164,  1.1742, -0.5951],
...0.7944, -0.3736,  0.8646, -1.1434,  0.7375,
          1.7932, -0.2706, -1.6484]], device='cuda:0', dtype=torch.float64)
        xmax       = tensor([[1.1742],
        [1.7951],
        [1.6296],
        [2.3410],
        [2.1546],
        [3.0498],
        [2.5522],
        [1.8421],
        [1.8344],
        [1.7932]], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([[ 0.3581],
        [-0.0078],
        [-0.3707],
        [-0.1180],
        [ 0.1290],
        [ 0.3572],
        [ 0.1914],
        [-0.5043],
        [ 0.2233],
        [ 0.3845]], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([[-0.6217,  0.0971, -0.2275,  0.8215, -0.1614,  0.3729, -0.9149, -0.9267,
          1.2593,  1.7931, -0.0237],
...0.2114, -0.0292, -1.1500,  0.1988, -0.1751,
          0.1181, -1.2614,  0.7835]], device='cuda:0', dtype=torch.float64)
        ym         = tensor([[-0.7552, -0.0363, -0.3610,  0.6881, -0.2949,  0.2394, -1.0483, -1.0602,
          1.1259,  1.6597, -0.1572],
...0.2048,  0.3869, -0.7338,  0.6150,  0.2411,
          0.5343, -0.8452,  1.1996]], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([[1.6597],
        [1.5898],
        [2.0992],
        [1.4359],
        [1.4451],
        [2.7608],
        [1.6752],
        [1.2973],
        [1.9243],
        [1.7343]], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([[ 0.1335],
        [ 0.4965],
        [ 0.1229],
        [ 0.1015],
        [-0.2455],
        [ 0.3079],
        [ 0.3285],
        [ 0.3237],
        [-0.0884],
        [-0.4162]], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor([ 0.5595, -0.3820, -0.0221, -0.5843,  0.0358, -0.1136,  0.4268, -0.0320,
         0.1027, -0.2719], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor([ 0.5595, -0.3820, -0.0221, -0.5843,  0.0358, -0.1136,  0.4268, -0.0320,
         0.1027, -0.2719], device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor([ 0.5595, -0.3820, -0.0221, -0.5843,  0.0358, -0.1136,  0.4268, -0.0320,
         0.1027, -0.2719], device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor([ 0.5595, -0.3820, -0.0221, -0.5843,  0.0358, -0.1136,  0.4268, -0.0320,
         0.1027, -0.2719], device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor([ 0.5595, -0.3820, -0.0221, -0.5843,  0.0358, -0.1136,  0.4268, -0.0320,
         0.1027, -0.2719], device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor([ 0.5595, -0.3820, -0.0221, -0.5843,  0.0358, -0.1136,  0.4268, -0.0320,
         0.1027, -0.2719], device='cuda:0', dtype=torch.float64)
_____________________________________________________________________________ TestPearsonr.test_array_api[greater-None-torch] ______________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'greater'
        axis       = None
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd28fe450>
        x          = array([[-0.24431625, -0.84084623, -0.07589802, -0.11144373,  0.21578661,
         1.49903863,  1.42430243,  0.92018406....90672059,  0.66825463,
        -0.46341451,  0.22889851,  2.14067528,  0.24860339, -1.42909962,
        -0.16551465]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[ 1.55576357,  1.28305173, -0.24697905,  1.17170251,  0.80859332,
         0.19826363,  0.22612805, -0.38215114....87020054, -0.37442017,
         0.77160499,  0.71135725,  1.06468037,  0.31461502, -2.02155495,
        -0.00855653]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 54.0
        alternative = 'greater'
        axis       = -1
        axis_int   = -1
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9d9fa9b010>
        dtype      = torch.float64
        method     = None
        n          = 110
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([9.9752], device='cuda:0', dtype=torch.float64)
        normym     = tensor([9.8779], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.0192, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([-0.2443, -0.8408, -0.0759, -0.1114,  0.2158,  1.4990,  1.4243,  0.9202,
         1.2874,  1.0433,  0.6790, -0....7,  0.6683,
        -0.4634,  0.2289,  2.1407,  0.2486, -1.4291, -0.1655], device='cuda:0',
       dtype=torch.float64)
        xm         = tensor([-0.3671, -0.9637, -0.1987, -0.2343,  0.0930,  1.3762,  1.3015,  0.7974,
         1.1646,  0.9205,  0.5562, -0....5,  0.5454,
        -0.5862,  0.1061,  2.0179,  0.1258, -1.5519, -0.2883], device='cuda:0',
       dtype=torch.float64)
        xmax       = tensor([3.6606], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([0.1228], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([ 1.5558,  1.2831, -0.2470,  1.1717,  0.8086,  0.1983,  0.2261, -0.3822,
        -0.6844,  0.0658, -0.2803, -0....2, -0.3744,
         0.7716,  0.7114,  1.0647,  0.3146, -2.0216, -0.0086], device='cuda:0',
       dtype=torch.float64)
        ym         = tensor([ 1.6093,  1.3366, -0.1934,  1.2253,  0.8621,  0.2518,  0.2797, -0.3286,
        -0.6308,  0.1193, -0.2268, -0....7, -0.3209,
         0.8252,  0.7649,  1.1182,  0.3682, -1.9680,  0.0450], device='cuda:0',
       dtype=torch.float64)
        ymax       = tensor([2.1887], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([-0.0535], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.0192, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.0192, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.0192, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.0192, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.0192, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.0192, device='cuda:0', dtype=torch.float64)
______________________________________________________________________________ TestPearsonr.test_array_api[two-sided-0-torch] ______________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'two-sided'
        axis       = 0
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd28feb90>
        x          = array([[-0.81837865, -0.39824587, -0.12878998,  0.99050264, -0.49923239,
        -0.51743688,  1.19669591,  0.25104772....96482996, -1.15333051,
         0.93607518,  0.78376363, -0.89119265, -0.32770967, -0.92657213,
        -0.47194279]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[ 0.88872796,  1.15057223,  0.17377089, -1.02033723,  0.00874263,
        -0.58315336,  2.09413673, -0.3772716 ....07241644,  1.05453301,
         1.25245848, -1.07471837,  0.44136057,  0.03251674,  0.8716997 ,
        -0.64639638]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 4.0
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 0
        const_x    = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        const_xy   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        const_y    = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd3b16910>
        dtype      = torch.float64
        method     = None
        n          = 10
        nconst_x   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        nconst_xy  = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        nconst_y   = tensor([False, False, False, False, False, False, False, False, False, False,
        False], device='cuda:0')
        normxm     = tensor([[2.8779],
        [2.6156],
        [2.8723],
        [1.7242],
        [1.6967],
        [2.7431],
        [3.3188],
        [3.8235],
        [2.0598],
        [2.2692],
        [2.6648]], device='cuda:0', dtype=torch.float64)
        normym     = tensor([[2.0415],
        [1.6744],
        [2.7180],
        [4.0000],
        [3.2600],
        [2.5410],
        [3.0668],
        [1.8488],
        [2.0288],
        [2.7120],
        [2.0513]], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor([ 0.2569, -0.1512, -0.3756,  0.1556, -0.3156,  0.5766,  0.0933, -0.1318,
         0.2480, -0.0556, -0.1262], device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([[-0.8184, -1.1140,  0.8830, -1.6956, -0.1717,  0.3178,  0.4442, -0.4181,
         -0.3253,  1.5163],
        [...0.1001, -1.6024,  0.7419, -0.7230, -0.9215,  0.5873,
         -1.0420, -0.4719]], device='cuda:0', dtype=torch.float64)
        xm         = tensor([[-6.8021e-01, -9.7583e-01,  1.0212e+00, -1.5574e+00, -3.3536e-02,
          4.5601e-01,  5.8241e-01, -2.7994e-...        -5.4714e-01, -7.4565e-01,  7.6319e-01, -8.6615e-01, -2.9605e-01]],
       device='cuda:0', dtype=torch.float64)
        xmax       = tensor([[1.6544],
        [1.5043],
        [1.3681],
        [0.9175],
        [1.0684],
        [1.5895],
        [2.1368],
        [1.8627],
        [1.2631],
        [1.1018],
        [1.4265]], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([[-0.1382],
        [-0.1494],
        [ 0.2112],
        [ 0.4735],
        [-0.5166],
        [-0.1564],
    ...89],
        [-0.3737],
        [ 0.3873],
        [-0.1645],
        [-0.1759]], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([[ 0.8887,  0.2932, -0.4813, -1.2553, -0.4926,  0.1203,  0.8482, -0.5886,
         -0.5360,  0.0642],
        [...1.3312,  0.2100, -0.0282,  0.4930,  0.8388,  0.0129,
         -0.2765, -0.6464]], device='cuda:0', dtype=torch.float64)
        ym         = tensor([[ 1.0026,  0.4071, -0.3674, -1.1414, -0.3787,  0.2342,  0.9622, -0.4746,
         -0.4221,  0.1781],
        [...1.2919,  0.2493,  0.0111,  0.5322,  0.8781,  0.0521,
         -0.2372, -0.6071]], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([[1.1414],
        [0.8536],
        [1.5278],
        [1.9165],
        [2.1754],
        [1.2461],
        [1.7633],
        [1.2179],
        [1.3204],
        [1.3186],
        [1.2919]], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([[-0.1139],
        [ 0.5515],
        [ 0.1106],
        [ 0.4347],
        [-0.3421],
        [ 0.6630],
    ...08],
        [ 0.1139],
        [ 0.6886],
        [-0.1453],
        [-0.0393]], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor([ 0.2569, -0.1512, -0.3756,  0.1556, -0.3156,  0.5766,  0.0933, -0.1318,
         0.2480, -0.0556, -0.1262], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor([ 0.2569, -0.1512, -0.3756,  0.1556, -0.3156,  0.5766,  0.0933, -0.1318,
         0.2480, -0.0556, -0.1262], device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor([ 0.2569, -0.1512, -0.3756,  0.1556, -0.3156,  0.5766,  0.0933, -0.1318,
         0.2480, -0.0556, -0.1262], device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor([ 0.2569, -0.1512, -0.3756,  0.1556, -0.3156,  0.5766,  0.0933, -0.1318,
         0.2480, -0.0556, -0.1262], device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor([ 0.2569, -0.1512, -0.3756,  0.1556, -0.3156,  0.5766,  0.0933, -0.1318,
         0.2480, -0.0556, -0.1262], device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor([ 0.2569, -0.1512, -0.3756,  0.1556, -0.3156,  0.5766,  0.0933, -0.1318,
         0.2480, -0.0556, -0.1262], device='cuda:0', dtype=torch.float64)
______________________________________________________________________________ TestPearsonr.test_array_api[two-sided-1-torch] ______________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'two-sided'
        axis       = 1
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd3cb1dd0>
        x          = array([[ 1.68424768, -1.56085057, -1.0802149 ,  1.04969255, -0.60140003,
        -0.08686482, -1.61103924, -0.60061338....05141612, -0.3521313 ,
         0.15033173,  2.37895364, -0.32511494, -0.18664921, -0.15406169,
        -0.08072992]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[-1.57439763, -0.57641921,  0.31417869,  0.89221701, -0.15176559,
        -1.54142646, -0.13442521,  1.00063607....63461387,  0.64302578,
         0.4761511 , -0.65656416,  1.02606876,  1.187437  ,  0.42534095,
        -0.04356603]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 4.5
        alternative = 'two-sided'
        axis       = -1
        axis_int   = 1
        const_x    = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        const_xy   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        const_y    = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd3bc1c10>
        dtype      = torch.float64
        method     = None
        n          = 11
        nconst_x   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        nconst_xy  = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        nconst_y   = tensor([False, False, False, False, False, False, False, False, False, False],
       device='cuda:0')
        normxm     = tensor([[3.2984],
        [3.0267],
        [2.8678],
        [2.3997],
        [2.7886],
        [3.9931],
        [4.0369],
        [2.5512],
        [4.3038],
        [3.3412]], device='cuda:0', dtype=torch.float64)
        normym     = tensor([[3.1455],
        [3.9349],
        [2.6807],
        [3.1961],
        [2.4867],
        [3.0249],
        [3.0043],
        [2.8816],
        [2.5340],
        [2.0497]], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor([-0.0241,  0.4226, -0.0329,  0.4160,  0.0671, -0.1617,  0.3120,  0.1541,
         0.1897, -0.2082], device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([[ 1.6842, -1.5609, -1.0802,  1.0497, -0.6014, -0.0869, -1.6110, -0.6006,
          0.0461, -0.6033,  0.5887],
...1.0514, -0.3521,  0.1503,  2.3790, -0.3251,
         -0.1866, -0.1541, -0.0807]], device='cuda:0', dtype=torch.float64)
        xm         = tensor([[ 1.9366, -1.3085, -0.8279,  1.3020, -0.3491,  0.1655, -1.3587, -0.3483,
          0.2984, -0.3510,  0.8410],
...1.3147, -0.6155, -0.1130,  2.1156, -0.5884,
         -0.4500, -0.4174, -0.3440]], device='cuda:0', dtype=torch.float64)
        xmax       = tensor([[1.9366],
        [1.7248],
        [1.9364],
        [1.1758],
        [1.3934],
        [2.6229],
        [2.6673],
        [1.6188],
        [2.9447],
        [2.1156]], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([[-0.2523],
        [ 0.5429],
        [ 0.0399],
        [-0.1504],
        [ 0.1753],
        [ 0.4399],
        [ 0.0073],
        [ 0.3410],
        [-0.0093],
        [ 0.2633]], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([[-1.5744, -0.5764,  0.3142,  0.8922, -0.1518, -1.5414, -0.1344,  1.0006,
          0.0072, -1.0858,  1.3409],
...0.6346,  0.6430,  0.4762, -0.6566,  1.0261,
          1.1874,  0.4253, -0.0436]], device='cuda:0', dtype=torch.float64)
        ym         = tensor([[-1.4372e+00, -4.3923e-01,  4.5137e-01,  1.0294e+00, -1.4578e-02,
         -1.4042e+00,  2.7623e-03,  1.1378e+...e-02, -1.1085e+00,  5.7415e-01,  7.3552e-01, -2.6581e-02,
         -4.9549e-01]], device='cuda:0', dtype=torch.float64)
        ymax       = tensor([[1.4781],
        [3.4099],
        [1.7523],
        [1.6137],
        [1.2680],
        [1.9101],
        [1.8024],
        [1.5532],
        [1.4155],
        [1.1085]], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([[-0.1372],
        [ 0.0094],
        [ 0.4703],
        [-0.3975],
        [-0.3548],
        [-0.3402],
        [ 0.0293],
        [-0.3803],
        [ 0.3129],
        [ 0.4519]], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor([-0.0241,  0.4226, -0.0329,  0.4160,  0.0671, -0.1617,  0.3120,  0.1541,
         0.1897, -0.2082], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor([-0.0241,  0.4226, -0.0329,  0.4160,  0.0671, -0.1617,  0.3120,  0.1541,
         0.1897, -0.2082], device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor([-0.0241,  0.4226, -0.0329,  0.4160,  0.0671, -0.1617,  0.3120,  0.1541,
         0.1897, -0.2082], device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor([-0.0241,  0.4226, -0.0329,  0.4160,  0.0671, -0.1617,  0.3120,  0.1541,
         0.1897, -0.2082], device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor([-0.0241,  0.4226, -0.0329,  0.4160,  0.0671, -0.1617,  0.3120,  0.1541,
         0.1897, -0.2082], device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor([-0.0241,  0.4226, -0.0329,  0.4160,  0.0671, -0.1617,  0.3120,  0.1541,
         0.1897, -0.2082], device='cuda:0', dtype=torch.float64)
____________________________________________________________________________ TestPearsonr.test_array_api[two-sided-None-torch] _____________________________________________________________________________
scipy/stats/tests/test_stats.py:681: in test_array_api
    res = stats.pearsonr(xp.asarray(x), xp.asarray(y),
        alternative = 'two-sided'
        axis       = None
        self       = <scipy.stats.tests.test_stats.TestPearsonr object at 0x7a9dd3cb0ad0>
        x          = array([[-0.22639395, -1.06346757, -0.50398259, -2.09876581,  0.8493841 ,
        -0.10586449, -0.51912194,  0.52055335....60613993,  0.76467161,
        -0.73527206, -0.52019479, -0.56367561,  0.39934576, -2.32879612,
        -1.84174229]])
        xp         = <module 'torch' from '/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/__init__.py'>
        y          = array([[-0.59088182,  0.487906  , -0.54788892, -0.05627702,  0.2122978 ,
        -0.49114164, -0.33629944,  1.34054938....54446181, -1.22667687,
         0.06588798,  0.37407459, -3.07566835,  0.78261166, -0.22503945,
         0.12370457]])
scipy/stats/_stats_py.py:4909: in pearsonr
    pvalue = _get_pvalue(np.asarray(r), dist, alternative)
        ab         = 54.0
        alternative = 'two-sided'
        axis       = -1
        axis_int   = -1
        const_x    = tensor(False, device='cuda:0')
        const_xy   = tensor(False, device='cuda:0')
        const_y    = tensor(False, device='cuda:0')
        dist       = <scipy.stats._distn_infrastructure.rv_continuous_frozen object at 0x7a9dd283a790>
        dtype      = torch.float64
        method     = None
        n          = 110
        nconst_x   = tensor(False, device='cuda:0')
        nconst_xy  = tensor(False, device='cuda:0')
        nconst_y   = tensor(False, device='cuda:0')
        normxm     = tensor([10.6620], device='cuda:0', dtype=torch.float64)
        normym     = tensor([10.6182], device='cuda:0', dtype=torch.float64)
        one        = tensor(1., device='cuda:0', dtype=torch.float64)
        r          = tensor(0.0308, device='cuda:0', dtype=torch.float64)
        threshold  = 1.8189894035458565e-12
        x          = tensor([-0.2264, -1.0635, -0.5040, -2.0988,  0.8494, -0.1059, -0.5191,  0.5206,
        -1.4072, -1.5916,  0.8970, -0....1,  0.7647,
        -0.7353, -0.5202, -0.5637,  0.3993, -2.3288, -1.8417], device='cuda:0',
       dtype=torch.float64)
        xm         = tensor([-0.1605, -0.9975, -0.4381, -2.0328,  0.9153, -0.0399, -0.4532,  0.5865,
        -1.3412, -1.5257,  0.9629,  0....2,  0.8306,
        -0.6693, -0.4543, -0.4977,  0.4653, -2.2629, -1.7758], device='cuda:0',
       dtype=torch.float64)
        xmax       = tensor([2.5180], device='cuda:0', dtype=torch.float64)
        xmean      = tensor([-0.0659], device='cuda:0', dtype=torch.float64)
        xp         = <module 'scipy._lib.array_api_compat.torch' from '/home/lucas/dev/myscipy/build-install/lib/python3.11/site-packages/scipy/_lib/array_api_compat/torch/__init__.py'>
        y          = tensor([-0.5909,  0.4879, -0.5479, -0.0563,  0.2123, -0.4911, -0.3363,  1.3405,
        -0.3880, -1.4808,  1.6382, -0....5, -1.2267,
         0.0659,  0.3741, -3.0757,  0.7826, -0.2250,  0.1237], device='cuda:0',
       dtype=torch.float64)
        ym         = tensor([-0.5525,  0.5263, -0.5095, -0.0179,  0.2507, -0.4527, -0.2979,  1.3789,
        -0.3496, -1.4424,  1.6766, -0....9, -1.1883,
         0.1043,  0.4125, -3.0373,  0.8210, -0.1866,  0.1621], device='cuda:0',
       dtype=torch.float64)
        ymax       = tensor([3.5431], device='cuda:0', dtype=torch.float64)
        ymean      = tensor([-0.0384], device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1060: in __array__
    return handle_torch_function(Tensor.__array__, (self,), self, dtype=dtype)
        dtype      = None
        self       = tensor(0.0308, device='cuda:0', dtype=torch.float64)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/overrides.py:1604: in handle_torch_function
    result = mode.__torch_function__(public_api, types, args, kwargs)
        args       = (tensor(0.0308, device='cuda:0', dtype=torch.float64),)
        kwargs     = {'dtype': None}
        mode       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        overloaded_args = [tensor(0.0308, device='cuda:0', dtype=torch.float64)]
        public_api = <function Tensor.__array__ at 0x7a9e75bc32e0>
        relevant_args = (tensor(0.0308, device='cuda:0', dtype=torch.float64),)
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/utils/_device.py:77: in __torch_function__
    return func(*args, **kwargs)
        args       = (tensor(0.0308, device='cuda:0', dtype=torch.float64),)
        func       = <function Tensor.__array__ at 0x7a9e75bc32e0>
        kwargs     = {'dtype': None}
        self       = <torch.utils._device.DeviceContext object at 0x7a9dd41db790>
        types      = (<class 'torch.Tensor'>,)
/home/lucas/mambaforge/envs/scipy-dev/lib/python3.11/site-packages/torch/_tensor.py:1062: in __array__
    return self.numpy()
E   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
        dtype      = None
        self       = tensor(0.0308, device='cuda:0', dtype=torch.float64)
================================================================================================ XFAILURES =================================================================================================
________________________________________________________________________ Test_ttest_CI.test_confidence_interval[0.2-True-two-sided] ________________________________________________________________________
scipy/stats/tests/test_stats.py:5422: in test_confidence_interval
    pytest.xfail('Discrepancy in `main`; needs further investigation.')
E   _pytest.outcomes.XFailed: Discrepancy in `main`; needs further investigation.
        alternative = 'two-sided'
        equal_var  = True
        self       = <scipy.stats.tests.test_stats.Test_ttest_CI object at 0x7a9dd275ffd0>
        trim       = 0.2
__________________________________________________________________________ Test_ttest_CI.test_confidence_interval[0.2-True-less] ___________________________________________________________________________
scipy/stats/tests/test_stats.py:5422: in test_confidence_interval
    pytest.xfail('Discrepancy in `main`; needs further investigation.')
E   _pytest.outcomes.XFailed: Discrepancy in `main`; needs further investigation.
        alternative = 'less'
        equal_var  = True
        self       = <scipy.stats.tests.test_stats.Test_ttest_CI object at 0x7a9dd275f110>
        trim       = 0.2
_________________________________________________________________________ Test_ttest_CI.test_confidence_interval[0.2-True-greater] _________________________________________________________________________
scipy/stats/tests/test_stats.py:5422: in test_confidence_interval
    pytest.xfail('Discrepancy in `main`; needs further investigation.')
E   _pytest.outcomes.XFailed: Discrepancy in `main`; needs further investigation.
        alternative = 'greater'
        equal_var  = True
        self       = <scipy.stats.tests.test_stats.Test_ttest_CI object at 0x7a9dd275f3d0>
        trim       = 0.2

mdhaber · 2024-03-26T00:44:30Z

Didn't we expect this given use of np.asarray to compute p-values?

lucascolley · 2024-03-26T00:48:14Z

Didn't we expect this given use of np.asarray to compute p-values?

Yep, I think it just means that we need to @skip_xp_backends(cpu_only=True).

[skip ci]

lucascolley

In an env with CuPy, python dev.py test -t scipy.stats.tests.test_stats -b all passes for me. LGTM once there is approval from the stats side!

tylerjereddy · 2024-03-26T20:33:18Z

I've not been following these closely, but have we changed our policy to allow array API support to be added partially to modules now?

lucascolley · 2024-03-26T20:42:08Z

I've not been following these closely, but have we changed our policy to allow array API support to be added partially to modules now?

Partial support is currently released in special, so I assume so.

tupui

The stats part is fine to me. And Lucas had a look at the Array API side so good to go. We could have moved the utils to the global utils that we have for Array API, though I don't think we will really "forget" so ok like this.

lucascolley · 2024-03-31T14:10:04Z

We can replace _move_axis_to_end and _clip when array_api_strict is updated.

Upstream issue: data-apis/array-api-strict#25

mdhaber added 5 commits February 22, 2024 09:09

WIP: stats.pearsonr: add support for axis argument

81491c2

TST: stats.pearsonr: add multidimensional tests, pass existing tests

16de34e

MAINT: stats.pearsonr: fix CI failures

2215dc4

DOC: stats.pearsonr: fix doctest

3bfc1c0

[skip actions] [skip cirrus]

Merge remote-tracking branch 'upstream/main' into gh9307

d05645b

mdhaber added scipy.stats array types Items related to array API support and input array validation (see gh-18286) labels Mar 19, 2024

lucascolley added the enhancement A new feature or improvement label Mar 20, 2024

mdhaber added 4 commits March 20, 2024 11:14

MAINT: stats.pearsonr: respect dtype

88a27ff

MAINT: stats.pearsonr: fixup

06c3758

MAINT: stats.pearsonr: scalar dtype fixup for statistic

5cbe4ab

stats.pearsonr: add Array API support

675dde6

mdhaber force-pushed the pearsonr_array_api branch from c5d60dd to 675dde6 Compare March 21, 2024 04:18

mdhaber changed the title ~~WIP: stats.pearsonr: add Array API support~~ ENH: stats.pearsonr: add Array API support Mar 21, 2024

lucascolley reviewed Mar 21, 2024

View reviewed changes

scipy/stats/_morestats.py Outdated Show resolved Hide resolved

scipy/stats/_stats_py.py Outdated Show resolved Hide resolved

scipy/stats/_stats_py.py Outdated Show resolved Hide resolved

mdhaber added 2 commits March 21, 2024 05:53

Merge remote-tracking branch 'upstream/main' into pearsonr_array_api

4ad786a

TST: stats.pearsonr: xfail Monte Carlo p-value test on 32 bit

6ddde92

[skip ci]

mdhaber force-pushed the pearsonr_array_api branch from 02242a4 to 6ddde92 Compare March 21, 2024 12:56

MAINT: stats.pearsonr: array API compatibility fixups

d19711a

mdhaber marked this pull request as ready for review March 21, 2024 13:03

mdhaber commented Mar 21, 2024

View reviewed changes

scipy/stats/_stats_py.py Outdated Show resolved Hide resolved

lucascolley reviewed Mar 21, 2024

View reviewed changes

scipy/stats/_stats_py.py Outdated Show resolved Hide resolved

scipy/stats/_stats_py.py Outdated Show resolved Hide resolved

mdhaber commented Mar 21, 2024

View reviewed changes

scipy/stats/_stats_py.py Show resolved Hide resolved

mdhaber commented Mar 21, 2024

View reviewed changes

TST: stats.pearsonr: make tests array API compatible

a493063

mdhaber commented Mar 22, 2024

View reviewed changes

scipy/stats/tests/test_stats.py Outdated Show resolved Hide resolved

Update scipy/stats/tests/test_stats.py

efd3cff

mdhaber commented Mar 22, 2024

View reviewed changes

.github/workflows/array_api.yml Outdated Show resolved Hide resolved

mdhaber mentioned this pull request Mar 24, 2024

MAINT, BUG (?): pearsonr statistic return type change #20324

Closed

lucascolley reviewed Mar 25, 2024

View reviewed changes

lucascolley changed the title ~~ENH: stats.pearsonr: add Array API support~~ ENH: stats.pearsonr: add array API support Mar 25, 2024

mdhaber added 3 commits March 25, 2024 20:19

MAINT: stats.pearsonr: test array API CPU only

10cdd9e

[skip ci]

Merge remote-tracking branch 'upstream/main' into pearsonr_array_api

9a5d4f5

MAINT: stats.pearsonr: rename skip_if_array_api -> skip_xp_backends

dedd975

lucascolley approved these changes Mar 26, 2024

View reviewed changes

lucascolley removed request for andyfaff and larsoner March 26, 2024 18:45

tylerjereddy mentioned this pull request Mar 27, 2024

WIP, ENH: signal: test_spectral array API support #19001

Draft

23 tasks

tupui approved these changes Mar 29, 2024

View reviewed changes

tupui merged commit 1066153 into scipy:main Mar 29, 2024

mdhaber mentioned this pull request Mar 30, 2024

RFC: SciPy array types & libraries support #18286

Open

mdhaber mentioned this pull request Apr 20, 2024

ENH: stats: add array API-support #20544

Open

mdhaber added this to the 1.14.0 milestone May 25, 2024

	@skip_if_array_api(np_only=True)
	@pytest.mark.parametrize("dtype", [np.float16, np.longdouble])
	def test_dtypes_nonstandard(self, dtype):
	x = random(30).astype(dtype)
	out_dtypes = {np.float16: np.complex64, np.longdouble: np.clongdouble}
	x_complex = x.astype(out_dtypes[dtype])

	res_fft = fft.ifft(fft.fft(x))
	res_rfft = fft.irfft(fft.rfft(x))
	res_hfft = fft.hfft(fft.ihfft(x), x.shape[0])
	# Check both numerical results and exact dtype matches
	assert_array_almost_equal(res_fft, x_complex)
	assert_array_almost_equal(res_rfft, x)
	assert_array_almost_equal(res_hfft, x)
	assert res_fft.dtype == x_complex.dtype
	assert res_rfft.dtype == np.result_type(np.float32, x.dtype)
	assert res_hfft.dtype == np.result_type(np.float32, x.dtype)

	@pytest.mark.parametrize("dtype", ["float32", "float64"])
	def test_dtypes_real(self, dtype, xp):
	x = xp.asarray(random(30), dtype=getattr(xp, dtype))

	res_rfft = fft.irfft(fft.rfft(x))
	res_hfft = fft.hfft(fft.ihfft(x), x.shape[0])
	# Check both numerical results and exact dtype matches
	rtol = {"float32": 1.2e-4, "float64": 1e-8}[dtype]
	xp_assert_close(res_rfft, x, rtol=rtol, atol=0)
	xp_assert_close(res_hfft, x, rtol=rtol, atol=0)

	@pytest.mark.parametrize("dtype", ["complex64", "complex128"])
	def test_dtypes_complex(self, dtype, xp):
	x = xp.asarray(random(30), dtype=getattr(xp, dtype))

	res_fft = fft.ifft(fft.fft(x))
	# Check both numerical results and exact dtype matches
	rtol = {"complex64": 1.2e-4, "complex128": 1e-8}[dtype]
	xp_assert_close(res_fft, x, rtol=rtol, atol=0)

Uh oh!

ENH: stats.pearsonr: add array API support #20284

ENH: stats.pearsonr: add array API support #20284

Uh oh!

Conversation

mdhaber commented Mar 19, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reference issue

What does this implement/fix?

Additional information

Uh oh!

lucascolley commented Mar 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mdhaber commented Mar 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lucascolley commented Mar 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lucascolley left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

lucascolley left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mdhaber Mar 21, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lucascolley Mar 21, 2024

Choose a reason for hiding this comment

Uh oh!

lucascolley Mar 21, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lucascolley Mar 21, 2024

Choose a reason for hiding this comment

Uh oh!

mdhaber Mar 21, 2024

Choose a reason for hiding this comment

Uh oh!

lucascolley Mar 21, 2024

Choose a reason for hiding this comment

Uh oh!

mdhaber Mar 21, 2024

Choose a reason for hiding this comment

Uh oh!

mdhaber Mar 22, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mdhaber commented Mar 21, 2024

Uh oh!

lucascolley commented Mar 21, 2024

Uh oh!

Uh oh!

mdhaber commented Mar 22, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lucascolley commented Mar 22, 2024

Uh oh!

Uh oh!

rgommers commented Mar 22, 2024

Uh oh!

lucascolley left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lucascolley Mar 25, 2024

Choose a reason for hiding this comment

Uh oh!

mdhaber commented Mar 19, 2024 •

edited

Loading

lucascolley commented Mar 20, 2024 •

edited

Loading

mdhaber commented Mar 20, 2024 •

edited

Loading

lucascolley commented Mar 20, 2024 •

edited

Loading

mdhaber Mar 21, 2024 •

edited

Loading

lucascolley Mar 21, 2024 •

edited

Loading

mdhaber Mar 22, 2024 •

edited

Loading

mdhaber commented Mar 22, 2024 •

edited

Loading

lucascolley left a comment •

edited

Loading