seed: 42

data:
  raw_dir: data/raw
  processed_dir: data/processed
  shard_size: 2000
  train_ratio: 0.8
  val_ratio: 0.1
  test_ratio: 0.1

features:
  use_degree: true
  use_log_degree: true
  use_normalised_degree: true
  use_degree_bucket: true
  degree_bucket_bins: 16
  use_clustering: true
  use_kcore: true
  use_pagerank: true
  use_laplacian_pe: true
  laplacian_pe_dim: 8
  use_rwse: true
  rwse_steps: 8
  pagerank_alpha: 0.85
  pagerank_max_iter: 100

model:
  architecture: gin
  hidden_dim: 128
  num_layers: 4
  dropout: 0.2
  num_heads: 4
  use_virtual_node: true
  pooling: attention
  gin_eps: 0.0
  pna_aggregators:
    - mean
    - max
    - min
    - std
  pna_scalers:
    - identity
    - amplification
    - attenuation

training:
  batch_size: 128
  num_epochs: 200
  learning_rate: 0.001
  weight_decay: 0.0001
  grad_clip: 1.0
  early_stopping_patience: 20
  scheduler: cosine
  use_amp: true
  num_workers: 2

output:
  runs_dir: runs
  checkpoints_dir: checkpoints