Experiments

experiments/ stores correctness, performance, and real-world study layouts. Each committed experiment keeps configuration and evaluation tooling in git; PLINK inputs and run outputs are generated locally and ignored.

experiments/
  correctness/
    tiny_even/
      README.md
      config.yaml
      configs/
        server/config.yaml
        center_1/config.yaml
        center_2/config.yaml
      data/
      results/
  performance/
    scales.yaml
    small_even/
    medium_even/
  real_world/
    phenotype_generation/
    1000genomes/
  tools/
    generate_baseline.py
    collect_run_metrics.py
    evaluation/

Scenario catalog

Scenario	Purpose	Clients	Status
correctness/tiny_even	Default two-center correctness validation	2	Configured
performance/small_even	Small scalability run	2	Configured
performance/medium_even	Medium scalability run	2	Configured
real_world/1000genomes	1000 Genomes chr22 application experiment	2	Configured

Generate synthetic data

python pipeline/simulation/simulated_data/generate_synthetic_data.py \
  --scale tiny \
  --partition-strategy even \
  --seed 42 \
  --output-dir experiments/correctness/tiny_even/data

The synthetic data generator lives in the local-only pipeline/simulation/ tree. It is referenced by the experiment docs, but generated PLINK data is not committed.

Run a scenario

flwr run . local-simulation --stream --run-config \
  'simulation=true num-server-rounds=100 config_path="experiments/correctness/tiny_even/configs"'

Result layout

Scenario results are organized under each scenario's configured results directory. The shipped tiny configs currently use results_2/:

• center_x/intermediate/: client intermediate files.
• center_x/logs/: client logs and PLINK outputs.
• server/logs/: server logs and stage metrics.
• server/intermediate/: temporary merged data, KING outputs, and LR outputs.
• Root-level merged metrics and retention manifests when monitoring and retention are enabled.