Update SVHN to create a dataset folder

[23pointsNorth]

Current behaviour:
torchvision.datasets.SVHN(root=' /some/dir/here', download=True) downloads the train_32x32.mat directly to root.

Issue:
It doesn't create an SVHN folder where the mat is stored, as the description of the root variable is.

Solution:
With this fix, the mat file will be downloaded in a root+'SVHN' folder and will be read from there. This way, the data is not placed in root without any label what the dataset is and is in a corresponding folder..

[23pointsNorth]

Can I have some information about the failed tests? Looking for details requires authorizing access to the bot.

[pmeier]

HI @23pointsNorth thanks for the PR!

While your fix aligns the functionality with the documentation, it breaks BC. I think it is more reasonable to change the documentation to actually state what is happening. Thoughts?

Looking for details requires authorizing access to the bot.

I've also struggled with this in the past. Strangely enough, you only have to be logged in at CircleCI to see the results. In the past I've just logged in with my GitHub and visited the link again to see the run.

Can I have some information about the failed tests?

The following test fails, since it tests the old behavior:

vision/test/test_datasets.py

Lines 211 to 223 in 6db1569

	@mock.patch('torchvision.datasets.SVHN._check_integrity')
	@unittest.skipIf(not HAS_SCIPY, "scipy unavailable")
	def test_svhn(self, mock_check):
	mock_check.return_value = True
	with svhn_root() as root:
	dataset = torchvision.datasets.SVHN(root, split="train")
	self.generic_classification_dataset_test(dataset, num_images=2)
	dataset = torchvision.datasets.SVHN(root, split="test")
	self.generic_classification_dataset_test(dataset, num_images=2)
	dataset = torchvision.datasets.SVHN(root, split="extra")
	self.generic_classification_dataset_test(dataset, num_images=2)

[fmassa]

I agree with @pmeier here, I think it would be preferable to change the documentation so that we keep backwards-compatibility.

[yassineAlouini]

I think the best thing to do here is to update the documentation as suggested by @pmeier. Should we close this PR in the time being or push the documentation update here? 🤔

[pmeier]

I would close the PR and go for the documentation change. Would you be up for that @yassineAlouini?

[yassineAlouini]

I would close the PR and go for the documentation change. Would you be up for that @yassineAlouini?

Yes, that works for me. 👌

[23pointsNorth]

Closing this PR and adding just the docs change in #5930.