seed: 2019
data:
  path: data/meteonorm_rf_hourly.npz
  synthetic_days: 30
  start_day_of_year: 1
  stations: 12
  pollutants: [PM2.5, PM10, NO2, SO2, O3, CO]
  frequency: hourly
  train_fraction: 0.70
model:
  engineering:
    trees: 12
    max_depth: 8
    min_samples_leaf: 8
    max_features: sqrt
    split_candidates: 12
paper_protocol:
  trees: hundreds
  normalization_resamples: 1000
normalization:
  engineering_resamples: 32
  season_window_days: 14
  match_hour: true
  preserve_time_features: [ttrend, day_of_year, weekend, hour]
paths:
  checkpoint: result/checkpoints/meteonorm_rf.pkl
  training_metrics: result/training/metrics.json
  inference: result/output/predictions.npz
  evaluation_dir: result/evaluation