flcore

A simple, straightforward federated learning micro framework for scholar.

Features

• Extensible, straightforward and simple structure.
• Integrate some useful utilities for FL and data:
  • io.py: Atomic write so that no more afraid of interruption when writing.
  • data.py: Out-of-box data split method to generate dataset for FL.
  • model.py: Easy model parameters operation, scale gradient, map parameters etc.
  • robust.py: Out-of-box robust aggregation function.
• Beautiful progress bar and terminal logging with rich.
• Integrate tensorboard to monitor your experiment.
• LowMemoryClientMixin to save your cuda memory. (Not available now)
• Auto save state if keyboard interruption detected.
• Auto save log & metrics of all clients while system running.

Under work

• Fully test.
• Well document

How-tos

There is no tutorial for now since some API may change, but you can check example.py.

LICENSE

This project is open sourced and under GNU General Public License v3.0.