basic:
  dataset: PEMSD4
  device: cuda:0
  mode: train
  model: STAEFormer
  seed: 2023

data:
  batch_size: 64
  column_wise: false
  days_per_week: 7
  horizon: 12
  input_dim: 1
  lag: 12
  normalizer: std
  num_nodes: 307
  steps_per_day: 288
  test_ratio: 0.2
  val_ratio: 0.1

model:
  adaptive_embedding_dim: 80
  dow_embedding_dim: 24
  dropout: 0.1
  feed_forward_dim: 256
  in_steps: 12
  input_dim: 1
  input_embedding_dim: 24
  num_heads: 4
  num_layers: 3
  num_nodes: 307
  out_steps: 12
  output_dim: 1
  spatial_embedding_dim: 0
  steps_per_day: 288
  tod_embedding_dim: 24
  use_mixed_proj: true

train:
  batch_size: 64
  debug: false
  early_stop: true
  early_stop_patience: 30
  epochs: 200
  grad_norm: false
  log_step: 2000
  loss_func: Huber
  lr_decay: true
  lr_decay_rate: 0.1
  lr_decay_step: 5,20,40,70
  lr_init: 0.001
  mae_thresh:
  mape_thresh: 0.0
  output_dim: 1
  plot: false
  real_value: true
  seed: 10
  weight_decay: 0.0003