seed: 2020
data:
  root: data
  format_version: rf_climparam_v2
  scales: [x4, x8, x16, x32]
  snapshots_per_scale: 1
  train_columns_per_latitude: 1
  vertical_levels: 48
  low_level_diffusion_levels: 15
  snapshot_hours: 3
  training_days: 337.5
  high_resolution:
    resolution_km: 12
    grid: [576, 1440]
    domain_km: [6912, 17280]
  coarse_grids:
    x4: {factor: 4, resolution_km: 48, grid: [144, 360]}
    x8: {factor: 8, resolution_km: 96, grid: [72, 180]}
    x16: {factor: 16, resolution_km: 192, grid: [36, 90]}
    x32: {factor: 32, resolution_km: 384, grid: [18, 45], online_native_grid: [18, 48]}
model:
  rf_tend: {input_features: 145, output_features: 144}
  rf_diff: {input_features: 62, output_features: 17}
  engineering:
    trees: 2
    max_depth: 5
    min_samples_leaf: 3
    max_features: sqrt
    split_candidates: 8
    bootstrap: true
paper_model:
  trees: 10
  min_samples_leaf: {x4: 20, x8: 20, x16: 20, x32: 7}
  training_samples: {x4: 5000000, x8: 5000000, x16: 5000000, x32: not_reported}
  max_depth: not_reported
  max_features: not_reported
  bootstrap: not_reported
  split_criterion: not_reported
  implementation: scikit-learn-0.21.2-RandomForestRegressor
runtime:
  device: cpu
paths:
  checkpoint: result/checkpoints/rf_climparam.pt
  training_metrics: result/training/metrics.json
  inference: result/output/predictions.npz
  evaluation_dir: result/evaluation
evaluation:
  precipitation_seconds: 10800
  online_proxy_scale: x32
  online_proxy_grid: [18, 48]
  prediction_chunk_size: 4096