fed-playground

A modular, pure-NumPy federated learning simulation and benchmark framework for research and education. Swap any (model × aggregation × encryption × attack × partition) and read off the privacy / robustness / utility trade-off — no cluster, no heavy deps.

Install

uv sync                 # core
uv sync --extra examples  # + scikit-learn, tqdm (some demos / datasets)
pip install fed-playground      # once released

30-second tour

from fed_playground import Environment, KrumAggregation, SignFlipAttack

env = Environment(
    n_parties=11, n_features=5, n_samples=900,
    aggregation_strategy=KrumAggregation(n_byzantine=2),
    attack=SignFlipAttack(scale=10), n_byzantine=2,
)
print(env.run_simulation(rounds=8)["global_loss"][-1])  # robust under attack

Or sweep a whole matrix with one command:

fedbench run benchmarks/robustness.toml

Where to go next

• API reference — every public class, pulled from its docstring.
• Extending — add your own aggregator or attack in ~20 lines.
• Benchmark study — the reproducible privacy × robustness × utility results.

The framework is built on a strategy pattern: Model, AggregationStrategy, EncryptionScheme, Attack, and Visualizer are independent, swappable ABCs. Robust aggregators and masking encryption schemes carry an is_linear_only contract so incompatible combinations are refused, not silently miscomputed.