Version 0.1.4ΒΆ
Release Date: December 08, 2024
π OverviewΒΆ
Improves FedAvg implementation with proper weighting based on client dataset sizes, enhances model versioning, and fixes type-related issues.
π Detailed NotesΒΆ
Improved FedAvg implementation to properly weight client contributions based on their dataset sizes and save the client weights. Additionally, several type-related issues have been fixed.
π FeaturesΒΆ
Implemented proper dataset size-based weighting in FedAvg aggregator
Added client waits to training metrics
Enhanced model versioning with support for weighted metrics
π Bug FixesΒΆ
Fixed averaging in FedAvg to properly weight based on client dataset sizes
Fixed MNIST data loader type hints
Fixed model versioning integration with the server and coordinator
π DocumentationΒΆ
Updated the release process.
Updated the
ModelManagersection in Concepts Guide.
β»οΈ Code RefactoringΒΆ
Updated
ModelManagerintegration with server and coordinator components.Improved aggregator interface with proper weight computation abstraction.
Simplified checkpointing by removing redundant storage.
π§ͺ TestsΒΆ
Added tests for weighted FedAvg.
Updated test cases to verify proper client weight calculations.
Added test coverage for missing sample counts in metrics.
π¦ Build SystemΒΆ
Added an automated release script.
π§ MaintenanceΒΆ
Fixed type checking issues.
π ChangelogΒΆ
template date parsing 58d0742
mypy issue 284148c
mypy issue 55a7c08
add release command to Makefile and fix mypy error b78a29d
fix mypy issue in CI 7359568
update fedavg to properly weight clients ac2d50c
make model management less redundant 1920ffd
update model management concept 35378d6
add automated release back into changelog and prepare_relase f46fd19
add automated release f707247
prepare release v0.1.4 90a2f9d
Merge pull request #5 from camille-004/cd/fedavg-weights (6533fd1) ()
Merge pull request #4 from camille-004/model-management (a30e8e3) ()
Merge pull request #3 from camille-004/automate-release (1331db9) ()