Configuration

Each participating center has its own config.yaml. The template is stored at configs/config_template.yaml. The default run reads from experiments/correctness/tiny_even/configs, which contains server/config.yaml, center_1/config.yaml, and center_2/config.yaml.

Center configuration

input_data:
  path: "experiments/correctness/tiny_even/data/tiny/center_1/tiny_center_1"
  type: "bed"

output:
  intermediate_dir: "experiments/correctness/tiny_even/results_2/center_1/intermediate"
  log_dir: "experiments/correctness/tiny_even/results_2/center_1/logs"

parameters:
  sample_offset: 1000000000000
  chunk_size: 100
  sample_chunk_size: 100
  snp_chunk_size: 100
  lr_target_chunks: 2
  run_id: "20260221_2000"

thresholds:
  maf_threshold: 0.01
  missing_threshold: 0.01
  hwe_threshold: 1e-6
  p_threshold: 0.3
  local_lr_threshold: 0.3
  global_lr_threshold: 0.1
  king_threshold: 0.23

flower:
  server_address: "127.0.0.1:8080"
  num_rounds: 10

participation:
  key_exchange: true
  sync: true
  local_qc: true
  global_qc: true
  global_qc_response: true
  init_chunks: true
  iterative_king: true
  local_lr: true
  local_lr_filter_response: true
  init_chunks_lr: true
  iterative_lr: true

Field reference

Field	Purpose
input_data.path	PLINK dataset prefix. Do not include .bed, .bim, or .fam. For VCF input, use the VCF file path.
input_data.type	Input type. The loader currently handles bed and vcf.
output.intermediate_dir	Per-center directory for chunks and temporary files.
output.log_dir	Per-center directory for logs and PLINK outputs.
parameters.chunk_size	Default chunk size. KING commonly partitions by samples; final LR commonly partitions by SNPs.
parameters.lr_target_chunks	Optional target number of LR chunks; used by compute_lr_chunk_size.
parameters.run_id	Optional run tag used to scope intermediate files under run_<id>.
thresholds.maf_threshold	Minimum minor allele frequency used by global QC.
thresholds.missing_threshold	Maximum SNP missingness allowed by global QC.
thresholds.hwe_threshold	Hardy-Weinberg equilibrium p-value threshold.
thresholds.p_threshold	Backward-compatible LR threshold.
thresholds.local_lr_threshold	Local LR threshold for selecting insignificant SNPs to filter before final LR.
thresholds.global_lr_threshold	Global LR threshold for classifying final LR p-values on clients.
thresholds.king_threshold	Kinship threshold used by client-side KING filtering.
participation	Per-stage opt-in flags. Use true for all stages when running the full pipeline.

Server configuration

pipeline/server_app.py requires a server config file under the configured experiment path:

<config_path>/server/config.yaml

The server config declares the server output directories:

output:
  intermediate_dir: "experiments/correctness/tiny_even/results_2/server/intermediate"
  log_dir: "experiments/correctness/tiny_even/results_2/server/logs"

Flower application configuration

Flower entry points and default run settings live in pyproject.toml:

[tool.flwr.app.components]
serverapp = "pipeline.server_app:app"
clientapp = "pipeline.client_app:app"

[tool.flwr.app.config]
simulation = true
num-server-rounds = 300
default-partition-by = "samples"
config_path = "experiments/correctness/tiny_even/configs"

In simulation mode, pipeline/client_app.py derives the center config path from the partition id:

<config_path>/center_<partition_id + 1>/config.yaml