#6213 Prepend "meta" to MetaTensor.__repr__ and MetaTensor.__str__ for easier identification

[MathijsdeBoer]

Discussed in #6213

Description

Prepends "meta" to the MetaTensor.__repr__ and MetaTensor.__str__ output so printing a MetaTensor does not look the exact same as a regular torch.Tensor.

I don't expect this change to cause any breaks, with me running the risk of invoking xkcd 1172.

---

1 failure in ./runtests.sh -f -u --net --coverage:

======================================================================
FAIL: test_values (tests.test_tciadataset.TestTciaDataset)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "D:\PythonProjects\MONAI-SaveImageFormatting\tests\test_tciadataset.py", line 72, in test_values
    self.assertTrue(
AssertionError: False is not true

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.
• In-line docstrings updated.

[ericspod]

I believe we had decided to leave it as "tensor" to avoid any changes to appearances, did we have a more specific rationale @rijobro @Nic-Ma @wyli ?

[wyli]

I think pytorch has recently improved this for tensor subclasses:

>>> import torch
>>> class SubTensor(torch.Tensor):
...     pass
... 
>>> a = SubTensor([1.0])
>>> print(a)
SubTensor([1.])

[MathijsdeBoer]

@wyli Running pip freeze on the venv with which I'm making my changes contains torch==2.0.0, is this change to the torch.Tensor printing in an in-dev PyTorch version? I ask because I wasn't seeing this behaviour myself.

I've tried seeing what would happen if I would comment out MetaTensor.__repr__ and MetaTensor.__str__:

Python 3.10.10 (tags/v3.10.10:aad5f6a, Feb  7 2023, 17:20:36) [MSC v.1929 64 bit (AMD64)] on win32
>>> import monai
>>> mt = monai.data.MetaTensor([1.])
>>> print(mt)
Metadata
	affine: tensor([[1., 0., 0., 0.],
        [0., 1., 0., 0.],
        [0., 0., 1., 0.],
        [0., 0., 0., 1.]], dtype=torch.float64)
	space: RAS
Applied operations
[]

[wyli]

Hi @MathijsdeBoer my example in the previous comment is not about monai, it's a pytorch demo to show subclassing...

[MathijsdeBoer]

@wyli Right, but isn't MetaTensor already a subclass of torch.Tensor? In that case you would expect it to work like in your example, if I'm not mistaken.

[wyli]

these are from the MetaObj...

MONAI/monai/data/meta_obj.py

Lines 157 to 168 in d57bff6

	def __repr__(self) -> str:
	"""String representation of class."""
	out: str = "\nMetadata\n"
	if self.meta is not None:
	out += "".join(f"\t{k}: {v}\n" for k, v in self.meta.items())
	else:
	out += "None"
	out += "\nApplied operations\n"
	if self.applied_operations is not None:
	out += pprint.pformat(self.applied_operations, indent=2, compact=True, width=120)
	else:

that's why I think this PR is a good idea..

[wyli]

comment out MetaTensor.__repr__ and MetaTensor.__str__, and change this:
class MetaTensor(MetaObj, torch.Tensor):
to
class MetaTensor(torch.Tensor, MetaObj):
will get the __repr__ resolved to pytorch's implementation :)
but these changes will require more thorough tests

[MathijsdeBoer]

Right, order matters. I don't often work with multiple inheritance in my projects, so I didn't think of that! Thanks!

[wyli]

/build