Expose negate_key_metric in FedAvg recipes

[chesterxgchen]

Summary

• Add negate_key_metric pass-through to the base FedAvg recipe and framework-specific FedAvg recipe wrappers.
• Wire the flag to IntimeModelSelector so recipe users can select lower-is-better metrics such as losses without reporting a separate negated metric from clients.
• Add focused recipe tests for PyTorch and NumPy recipe pass-through.

Testing

• python -m pytest tests/unit_test/recipe/fedavg_recipe_test.py -q
• git diff --check

[chesterxgchen]

/build

[greptile-apps]

Greptile Summary

This PR adds a negate_key_metric: bool = False parameter to the entire FedAvg recipe hierarchy so users can select lower-is-better metrics (losses) without reporting a negated value from clients. The flag is threaded from each framework-specific recipe wrapper through the base FedAvgRecipe → BaseFedJob → IntimeModelSelector.

• Adds negate_key_metric to seven recipe classes (PyTorch, NumPy, sklearn, KMeans, SVM, TF, and the base recipe) and BaseFedJob, with consistent forwarding via super().__init__() in every wrapper.
• Wires the flag into IntimeModelSelector in BaseFedJob, which already supports the parameter — no changes to the selector itself were needed.
• Adds six new unit tests covering the full pass-through for every recipe variant, including a lightweight TF test that avoids a TensorFlow runtime dependency by lazily loading the module with a mock.

Confidence Score: 5/5

Safe to merge — the change is a straightforward bool parameter forwarded uniformly through all recipe wrappers to IntimeModelSelector, which already supports the flag.

All seven recipe wrappers consistently receive and forward negate_key_metric via super().init(). The base FedAvgRecipe validates and stores the flag, BaseFedJob threads it into IntimeModelSelector, and six new unit tests verify the full pass-through for every framework variant. No behavioral changes occur when the default False is used.

Files Needing Attention: No files require special attention.

Important Files Changed

Filename	Overview
nvflare/recipe/fedavg.py	Adds negate_key_metric to _FedAvgValidator, FedAvgRecipe.init, and BaseFedJob construction; forwarded correctly at all call sites.
nvflare/job_config/base_fed_job.py	Adds negate_key_metric param with correct docstring caveat; wired into IntimeModelSelector construction under the existing elif key_metric: guard.
nvflare/app_opt/pt/recipes/fedavg.py	Adds negate_key_metric pass-through to PyTorch FedAvgRecipe; docstring and super() call updated correctly.
nvflare/app_common/np/recipes/fedavg.py	Adds negate_key_metric pass-through to NumpyFedAvgRecipe; consistent with other recipe wrappers.
nvflare/app_opt/sklearn/recipes/fedavg.py	Adds negate_key_metric to SklearnFedAvgRecipe; correctly forwarded to UnifiedFedAvgRecipe super().init().
nvflare/app_opt/sklearn/recipes/kmeans.py	Adds negate_key_metric to KMeansFedAvgRecipe with appropriate docstring; correctly forwarded to base recipe.
nvflare/app_opt/sklearn/recipes/svm.py	Adds negate_key_metric to SVMFedAvgRecipe with appropriate docstring; correctly forwarded to base recipe.
nvflare/app_opt/tf/recipes/fedavg.py	Adds negate_key_metric to TF FedAvgRecipe; forwarded correctly via super().init() without any module-level TF imports that could break tests.
tests/unit_test/recipe/fedavg_recipe_test.py	Six new tests cover negate_key_metric pass-through for PT, NumPy, sklearn, KMeans, SVM, and TF recipes; TF test uses a clean module-loading approach to avoid TensorFlow dependency.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["Recipe user sets\nnegate_key_metric=True"] --> B["Framework-specific recipe\n(PT / NumPy / Sklearn / KMeans / SVM / TF)"]
    B --> C["Base FedAvgRecipe\nnvflare/recipe/fedavg.py"]
    C --> D["BaseFedJob\nnvflare/job_config/base_fed_job.py"]
    D -->|"key_metric provided\n(model_selector not supplied)"| E["IntimeModelSelector\n(key_metric, negate_key_metric)"]
    D -->|"custom model_selector supplied"| F["Custom FLComponent\n(negate_key_metric ignored)"]
    E --> G["Model comparison:\nvalue x -1 if negate_key_metric\nfor lower-is-better metrics"]

Loading

_{Reviews (5): Last reviewed commit: "Address negate metric recipe review comm..." | Re-trigger Greptile}

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 62.85%. Comparing base (d714126) to head (98a994f).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4953      +/-   ##
==========================================
+ Coverage   62.83%   62.85%   +0.02%     
==========================================
  Files         995      995              
  Lines       97868    97870       +2     
==========================================
+ Hits        61496    61520      +24     
+ Misses      36372    36350      -22     

Flag	Coverage Δ
unit-tests	62.85% <100.00%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[chesterxgchen]

/build

[chesterxgchen]

/build

[chesterxgchen]

/build