Update Transforms tutorial — replace deprecated ToTensor(), migrate to v2 namespace, modernize one-hot encoding

[sekyondaMeta]

Description

The Transforms tutorial uses the deprecated torchvision.transforms.ToTensor() transform, operates entirely in the legacy v1 namespace, and uses a low-level scatter_ pattern for one-hot encoding where F.one_hot is the modern alternative.

Changes needed

Deprecated APIs

Issue	Current Code	Replacement	Since
torchvision.transforms.ToTensor	from torchvision.transforms import ToTensor, Lambda / transform=ToTensor() (lines 31, 37)	from torchvision.transforms import v2 / transform=v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])	torchvision 0.16 (Oct 2023) — ToTensor emits a deprecation warning; v2 transforms are the supported API

Suboptimal / Outdated Patterns

Issue	Current Code	Modern Alternative	Notes
v1 transforms namespace throughout	from torchvision.transforms import ToTensor, Lambda	from torchvision.transforms import v2	The entire tutorial operates in the legacy namespace. All imports and explanations should shift to v2.
Lambda for one-hot encoding	Lambda(lambda y: torch.zeros(10, dtype=torch.float).scatter_(0, torch.tensor(y), value=1)) (line 38)	torch.nn.functional.one_hot(torch.tensor(y), num_classes=10).float()	F.one_hot is cleaner and more readable. Lambda itself still works in v2, but the one-hot pattern has a better built-in.
scatter_ for one-hot encoding	torch.zeros(10, dtype=torch.float).scatter_(dim=0, index=torch.tensor(y), value=1) (lines 60-61)	torch.nn.functional.one_hot(torch.tensor(y), num_classes=10).float()	scatter_ is a low-level op; F.one_hot is purpose-built and self-documenting.

Specific lines

• Line 31: from torchvision.transforms import ToTensor, Lambda -> from torchvision.transforms import v2
• Line 37: transform=ToTensor(), -> transform=v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)]),
• Line 38: target_transform=Lambda(lambda y: torch.zeros(10, dtype=torch.float).scatter_(0, torch.tensor(y), value=1)) -> target_transform=v2.Lambda(lambda y: torch.nn.functional.one_hot(torch.tensor(y), num_classes=10).float())
• Lines 42-48: Update the ToTensor() section heading and explanation to describe v2.ToImage() + v2.ToDtype()
• Lines 60-61: target_transform = Lambda(lambda y: torch.zeros(10, dtype=torch.float).scatter_(dim=0, index=torch.tensor(y), value=1)) -> target_transform = v2.Lambda(lambda y: torch.nn.functional.one_hot(torch.tensor(y), num_classes=10).float())

Files

• beginner_source/basics/transforms_tutorial.py

---

Copied from sekyondaMeta#71

cc @subramen

[omkar-334]

/assigntome