feat(examples): add richer evaluation metrics to advanced-pytorch

[SalimELMARDI]

Issue

Description

The advanced-pytorch example currently reports only basic evaluation metrics (loss/accuracy on server-side and eval_loss/eval_acc on client-side).
This limits visibility into ranking quality and class-level behavior during federated runs.

Related issues/PRs

Supersedes the earlier quickstart-focused PR : #6638 (moved to advanced-pytorch based on @chongshenng feedback).

Proposal

Explanation

This PR extends evaluation reporting in examples/advanced-pytorch while keeping existing metric keys backward-compatible.

Changes:

• Updated examples/advanced-pytorch/pytorch_example/task.py:
  • Extended test(...) to compute:
    • top-1 accuracy (existing behavior)
    • top-3 accuracy
    • per-class top-1 accuracy for Fashion-MNIST (class_accuracy_0 ... class_accuracy_9)
  • Uses torch.bincount-based accumulation for per-class stats.
• Updated examples/advanced-pytorch/pytorch_example/client_app.py:
  • Kept existing eval_loss and eval_acc
  • Added eval_acc_top3
  • Added eval_acc_class_0 ... eval_acc_class_9
• Updated examples/advanced-pytorch/pytorch_example/server_app.py:
  • Kept existing loss and accuracy
  • Added accuracy_top3
  • Added accuracy_class_0 ... accuracy_class_9

Validation:

• Ran a 1-round simulation locally:
  • flwr run .\examples\advanced-pytorch --stream --run-config "num-server-rounds=1 fraction-train=0.2 fraction-evaluate=0.2"
• 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 preserved for compatibility.

[Copilot]

Pull request overview

This PR enhances the examples/advanced-pytorch evaluation reporting so federated runs expose richer quality signals (top-k and per-class accuracy) while keeping existing metric keys intact (accuracy/loss server-side, eval_acc/eval_loss client-side).

Changes:

• Extend test(...) to compute top-3 accuracy and per-class top-1 accuracy (Fashion-MNIST, 10 classes).
• Emit the new metrics from both ClientApp.evaluate and server-side centralized evaluation.
• Introduce a shared NUM_CLASSES constant to drive per-class metric generation.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
examples/advanced-pytorch/pytorch_example/task.py	Computes top-3 and per-class accuracies and returns a metrics dict from test(...).
examples/advanced-pytorch/pytorch_example/client_app.py	Adds eval_acc_top3 and eval_acc_class_{i} metrics while preserving existing keys.
examples/advanced-pytorch/pytorch_example/server_app.py	Adds accuracy_top3 and accuracy_class_{i} metrics while preserving existing keys.

---

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