Last edited

Flower Datasets

Flower Datasets is a library by the Flower Labs team that helps create datasets for federated learning, analytics, and evaluation. It supports both naturally partitioned datasets and simulated heterogeneity from centralized data.

Key features:

  • Download datasets (powered by HuggingFace datasets)
  • Partition data with built-in or custom schemes
  • Generate centralized versions
  • Ensure reproducible experiments
  • Visualize partitions
  • Framework-agnostic for easy ML integration

It's designed to complement and integrate seamlessly with Flower.