feat(examples): Add top-k and per-class eval metrics to quickstart-pytorch

[SalimELMARDI]

Issue

Description

The quickstart-pytorch example only reported basic evaluation metrics (loss and top-1 accuracy).
That made it harder to inspect ranking quality and class-level performance during federated runs.

Related issues/PRs

N/A

Proposal

Explanation

This PR extends evaluation reporting in examples/quickstart-pytorch while keeping existing metrics backward-compatible.

Changes:

• Updated pytorchexample/task.py:
  • Extended test(...) to compute:
    • top-1 accuracy (existing behavior)
    • top-3 accuracy
    • per-class top-1 accuracy for CIFAR-10 (class_accuracy_0 ... class_accuracy_9)
• Updated pytorchexample/client_app.py:
  • Kept existing eval_loss and eval_acc
  • Added eval_acc_top3
  • Added eval_acc_class_0 ... eval_acc_class_9
• Updated pytorchexample/server_app.py:
  • Kept existing loss and accuracy
  • Added accuracy_top3
  • Added accuracy_class_0 ... accuracy_class_9
• Updated examples/quickstart-pytorch/README.md:
  • Documented the new reported metrics

Validation:

• Ran a 1-round simulation locally:
  • flwr run . --stream --run-config "num-server-rounds=1 batch-size=128 fraction-evaluate=0.1"
• Confirmed new client and server metrics are present in logs.

Checklist

• Implement proposed change
• Write tests
• Update documentation
• Make CI checks pass
• Ping maintainers on Slack (channel #contributions)

Any other comments?

No API-breaking changes. Existing metric keys were kept for compatibility.

[Copilot]

Pull request overview

This PR enhances the examples/quickstart-pytorch evaluation reporting to include top-3 accuracy and per-class (CIFAR-10) top-1 accuracies, exposing these metrics from both client-side evaluation and centralized server-side evaluation while keeping existing metric keys.

Changes:

• Extended test(...) to compute top-3 accuracy and per-class top-1 accuracies alongside existing loss/top-1 accuracy.
• Updated client and server apps to emit the additional metrics under new, backward-compatible metric keys.
• Documented the newly reported metrics in the example README.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.

File	Description
examples/quickstart-pytorch/pytorchexample/task.py	Computes top-3 and per-class accuracies and returns a metrics dict from test().
examples/quickstart-pytorch/pytorchexample/client_app.py	Adds client-side metric keys for top-3 and per-class accuracies.
examples/quickstart-pytorch/pytorchexample/server_app.py	Adds centralized metric keys for top-3 and per-class accuracies.
examples/quickstart-pytorch/README.md	Documents the expanded set of reported metrics (but front matter formatting changed).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[chongshenng]

Hello @SalimELMARDI, thanks for opening this PR. I do agree that it's a useful enhancement to our example PyTorch app. But I'm unsure if we should merge this in this form because we intentionally keep our quickstart apps simple - just run basic training and eval.

Maybe it's better to implement this in our advanced-pytorch example? Wdyt?

[SalimELMARDI]

Hello @SalimELMARDI, thanks for opening this PR. I do agree that it's a useful enhancement to our example PyTorch app. But I'm unsure if we should merge this in this form because we intentionally keep our quickstart apps simple - just run basic training and eval.

Maybe it's better to implement this in our advanced-pytorch example? Wdyt?

@chongshenng Thanks for the feedback, that makes sense. I’ll open a new PR with this enhancement in advanced-pytorch, then close this one as superseded.

[SalimELMARDI]

Author

Superseded by #6713