Modernize transforms tutorial to torchvision v2 API

[omkar-334]

Fixes #3853

Description

files changed - beginner_source/basics/transforms_tutorial.py

1. Replaced deprecated torchvision.transforms.ToTensor (deprecated since torchvision 0.16) with the v2 pipeline v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)]).
2. Switched the legacy v1 transforms namespace import to from torchvision.transforms import v2.
3. Replaced the low-level torch.zeros(...).scatter_(...) one-hot pattern with torch.nn.functional.one_hot(torch.tensor(y), num_classes=10).float(), wrapped in v2.Lambda.
4. Updated the section heading and explanatory prose so they describe v2.ToImage() + v2.ToDtype() and F.one_hot instead of ToTensor and scatter_.

Checklist

• The issue that is being fixed is referred in the description (see above "Fixes #ISSUE_NUMBER")
• Only one issue is addressed in this pull request
• Labels from the issue that this PR is fixing are added to this pull request
• No unnecessary issues are included into this pull request.

cc @subramen

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/tutorials/3861

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit 90cde13 with merge base 5efb99a ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[sekyondaMeta]

@claude

[claude]

Claude finished @sekyondaMeta's task in 1m 2s —— View job

---

Review of PR #3861

• Gather context and read changed files
• Review code changes for correctness and style
• Post review feedback

---

Overall: This is a clean, well-scoped modernization. The changes correctly migrate from the deprecated torchvision.transforms.ToTensor to the v2 API, and the one_hot replacement is more readable than the manual scatter_ pattern. A few observations:

Looks Good

1. v2 API usage is correct — v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)]) is the recommended replacement for ToTensor() per the torchvision docs.
2. one_hot replacement is cleaner — torch.nn.functional.one_hot(torch.tensor(y), num_classes=10).float() is more readable and idiomatic than torch.zeros(10).scatter_(0, torch.tensor(y), value=1).
3. Prose is clear and concise — the explanatory text matches the code and links to correct documentation pages.
4. Section heading updated appropriately — "ToImage() and ToDtype()" accurately describes the new content.

Minor Suggestions (non-blocking)

1. Line 38 is long (~100 chars) — Consider breaking the target_transform lambda across lines for readability, though this is a style preference and the current form is acceptable for a tutorial:

   target_transform=v2.Lambda(
       lambda y: torch.nn.functional.one_hot(torch.tensor(y), num_classes=10).float()
   )

   Same applies to line 61 where it's repeated.

2. import torch.nn.functional as F — The PR uses torch.nn.functional.one_hot inline (twice). A common pattern in PyTorch tutorials is import torch.nn.functional as F and then F.one_hot(...). This would shorten the lambda and match conventions used elsewhere in PyTorch docs. However, since this tutorial is minimal and doesn't use F for anything else, the fully qualified name is also fine.

3. Further Reading link — The "Further Reading" section at line 70 still links to the general transforms page. Consider adding or replacing with a direct link to the v2 transforms getting started guide for readers who want to explore the new API further.

Verdict

The PR is correct, well-motivated, and addresses issue #3853. The suggestions above are minor polish — this is ready to merge as-is.

---