The latest pytorch nightly has removed torch.testing.assert_allclose

[xuzhao9]

Describe the bug

The latest PyTorch nightly has removed torch.testing.assert_allclose:
pytorch/pytorch#87974

Kornia needs to update this code: https://github.com/kornia/kornia/blob/master/kornia/testing/__init__.py#L271

Reproduction steps

`import kornia`
on the latest PyTorch nightly

error:
Traceback (most recent call last):
  File "/home/circleci/miniconda3/lib/python3.8/site-packages/kornia/testing/__init__.py", line 235, in <module>
    from torch.testing._core import _get_default_tolerance  # type: ignore
ModuleNotFoundError: No module named 'torch.testing._core'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "install.py", line 20, in <module>
    patch_dalle2()
  File "install.py", line 7, in patch_dalle2
    import dalle2_pytorch
  File "/home/circleci/miniconda3/lib/python3.8/site-packages/dalle2_pytorch/__init__.py", line 2, in <module>
    from dalle2_pytorch.dalle2_pytorch import DALLE2, DiffusionPriorNetwork, DiffusionPrior, Unet, Decoder
  File "/home/circleci/miniconda3/lib/python3.8/site-packages/dalle2_pytorch/dalle2_pytorch.py", line 20, in <module>
    from kornia.filters import gaussian_blur2d
  File "/home/circleci/miniconda3/lib/python3.8/site-packages/kornia/__init__.py", line 3, in <module>
    from . import filters
  File "/home/circleci/miniconda3/lib/python3.8/site-packages/kornia/filters/__init__.py", line 2, in <module>
    from .blur_pool import BlurPool2D, MaxBlurPool2D, blur_pool2d, edge_aware_blur_pool2d, max_blur_pool2d
  File "/home/circleci/miniconda3/lib/python3.8/site-packages/kornia/filters/blur_pool.py", line 7, in <module>
    from kornia.testing import KORNIA_CHECK, KORNIA_CHECK_SHAPE
  File "/home/circleci/miniconda3/lib/python3.8/site-packages/kornia/testing/__init__.py", line 254, in <module>
    from torch.testing import assert_allclose as _assert_close
ImportError: cannot import name 'assert_allclose' from 'torch.testing' (/home/circleci/miniconda3/lib/python3.8/site-packages/torch/testing/__init__.py)

Expected behavior

successful return

Environment

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

• PyTorch Version (e.g., 1.0):
• OS (e.g., Linux):
• How you installed PyTorch (conda, pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information:

### Additional context

_No response_

[edgarriba]

@xuzhao9 thanks to report! we are aware of it. was recently pointed out in a conversation @johnnv1

[pmeier]

I actually authored the mitigation code in the kornia test suite. See #1031. The offending line here is

kornia/kornia/testing/__init__.py

Line 235 in a79fbcb

from torch.testing._core import _get_default_tolerance # type: ignore

Change this to

from torch.testing._comparison import default_tolerances

and you should be good again.

[edgarriba]

@pmeier do you think is safe to remove the try/catch block ? apparently this gives us also some headaches with mypy

[pmeier]

It will be after you bump the minimum dependency to torch>=1.9, but it seems that has not happened yet:

kornia/setup.cfg

Lines 38 to 40 in b9f26e0

	install_requires =
	packaging
	torch>=1.8.1

[edgarriba]

@ducha-aiki @shijianjian opinions here?

[pmeier]

Plus, my patch from #1031 only broke, because I needed to import from a private namespace. See #1134 for details. When that is fixed as well you can use torch.testing.assert_close without any workarounds I think.