💡 Learning

P2PFL offers a flexible and customizable learning process, allowing you to adapt your federated learning experiments to your specific requirements. The learning system is built on three core components:

• Learners: Manage local training, evaluation, and model updates on each node. They are framework-agnostic (supporting PyTorch, TensorFlow, Flax, etc.) and handle aggregator callbacks.

• Datasets: The P2PFLDataset class simplifies dataset loading, partitioning (IID, non-IID), and exporting for federated learning. It supports diverse data sources (CSV, JSON, Parquet, Pandas, Hugging Face, SQL).

• Aggregators: Combine model updates from different nodes using various strategies (FedAvg, FedMedian, SCAFFOLD, etc.) to create a new global model.

📋 Modules

• 🧠 Learners
• 🗂️ Datasets
• 🧩 Aggregators