Skip to content

Configuration: config/{iso}_config.yaml

Each country has a YAML configuration file that defines workflow parameters for analytical, trigger, and operational scripts. These files are read by the Params class in AA/params.py.

How It Works

  • The configuration file is loaded by: python config = load_config(iso) # reads ./config/{iso}_config.yaml
  • All keys in the YAML are mapped to attributes in Params.
  • Defaults (e.g., monitoring year, calibration year, paths) are applied if keys are missing.
  • Additional logic:
  • aggregate is set based on index (spi or dryspell).
  • districts are inferred from districts_vulnerability.
  • tolerance and requirements are converted to typed dictionaries for performance.
  • windows define indicator periods for each index family.

Main Sections in YAML

  • Identification
  • iso: Country code (e.g., moz, zwe).

  • index: Indicator family (spi or dryspell).

  • Season & Monitoring

  • issue_months: Months for trigger selection.
  • monitoring_year, calibration_year, start_monitoring.

  • start_season, end_season.

  • Vulnerability & Skill

  • vulnerability: GT, NRT, or TBD.
  • districts_vulnerability: Map of district → vulnerability class.
  • general_t and non_regret_t: Skill requirements for GT/NRT.

  • tolerance: Thresholds for false-alarm tolerance.

  • Windows

  • windows: Nested structure defining indicator periods for each index. Example: yaml windows: spi: window1: ["ON", "NDJ"] window2: ["DJF", "MAM"] dryspell: window1: ["NDJ"] window2: ["DJF"]

  • Paths

  • data_path and output_path: Where to read/write data.
  • Defaults to s3://wfp-ops-userdata/amine.barkaoui/aa if not provided.

Example Structure

iso: moz
index: spi
issue_months: [9, 10, 11]
monitoring_year: 2024
calibration_year: 2022
start_monitoring: 5

districts_vulnerability:
  Chokwe: GT
  Guija: GT
  Mandlakazi: NRT

general_t:
  hit_rate: 0.6
  success_rate: 0.55

non_regret_t:
  hit_rate: 0.5
  success_rate: 0.5

tolerance:
  mild: 0.05
  moderate: 0.10

windows:
  spi:
    window1: ["ON", "NDJ"]
    window2: ["DJF", "MAM"]
  dryspell:
    window1: ["NDJ"]
    window2: ["DJF"]

data_path: s3://wfp-ops-userdata/amine.barkaoui/aa
output_path: s3://wfp-ops-userdata/amine.barkaoui/aa

Key Points

  • Keep YAML minimal but complete for required sections.
  • Avoid hardcoding defaults in YAML unless overriding.
  • Validate changes against Params logic in params.py.
  • Document any country-specific deviations inline in the YAML.