I can get to the notebook. Which one is it? The colab link doesn't say where it is in the repo.

I'll update the tutorial https://colab.research.google.com/drive/1T4iAys-cC2qL80oJkIbAXAPlWNPwp4H7

/build

/build

The way tensors are printed was changed in Pytorch 1.12 which could be used to print "MetaTensor" instead of "tensor" but would need to be taken into account in the tests.

with this PR, print(metatensor) will output exactly the same string as print(tensor)...we lose all the meta info in the string, without this PR, print(metatensor) is verbose, any solution in the middle? cc @Nic-Ma @rijobro

Hi @ericspod @wyli ,

Actually, I still don't quite understand the target goal of this PR, usually the output of __str__ is for human reading, just in the format of a regular string, why not add some meta information in it?

Thanks.

Hi @ericspod @wyli ,

Actually, I still don't quite understand the target goal of this PR, usually the output of __str__ is for human reading, just in the format of a regular string, why not add some meta information in it?

Thanks.

Hi Nic, the issue is clearly described in #4840, the pr is trying to address 4840.

Hi @wyli ,

I checked the original ticket this morning, that's I don't quite understand...
Especially: "The design intent was to not require users to have to deal with MetaTensors, it's meant to be opt-in, but this change has a visible difference the unexpecting user would have to dig into to solve."
I think we want to highlight MetaTensor in MONAI release and let users to use it, why hide the meta information in the print?

Thanks.

The issue is that a lot of code will have print statements using tensor directly, with normal Pytorch tensors this isn't a problem but with MetaTensor it's way too verbose. It'll break some people's code I image who rely on output being in a specific format. It's just too surprising to use code that does known things in Pytorch then all of a sudden with MONAI produces a huge output with information that isn't likely relevant at the time. Users will be confused about where the metadata stuff came from, we want people to not have to deal with MetaTensor if they don't want but this forces them to do just that.

Hi @ericspod ,

OK, I see your point.
Could you please help add clear doc-string to the __str__ function and __repr__ function to let users or developers know the usage?

Thanks.

Hi @ericspod ,

OK, I see your point. Could you please help add clear doc-string to the __str__ function and __repr__ function to let users or developers know the usage?

Thanks.

Done.

/build