version: 0.0.1
pipeline_name: graphpred
device: cuda:0                # Torch device name, e.q. cpu or cuda or cuda:0
data:
  name: ogbg-molhiv
  split_ratio:                # Ratio to generate data split, for example set to [0.8, 0.1, 0.1] for 80% train/10% val/10% test. Leave blank to use builtin split in original dataset
model:
  name: pna
  embed_size: 80              # Embedding size.
  aggregators: mean max min std # Aggregation function names separated by space, can include mean, max, min, std, sum
  scalers: identity amplification attenuation # Scaler function names separated by space, can include identity, amplification, and attenuation
  dropout: 0.3                # Dropout rate.
  batch_norm: true            # Whether to use batch normalization.
  residual: true              # Whether to use residual connection.
  num_mlp_layers: 1           # Number of MLP layers to use after message aggregation in each PNA layer.
  num_layers: 4               # Number of PNA layers.
  readout: mean               # Readout for computing graph-level representations, can be 'sum' or 'mean'.
general_pipeline:
  num_runs: 10                # Number of experiments to run
  train_batch_size: 128       # Graph batch size when training
  eval_batch_size: 128        # Graph batch size when evaluating
  num_workers: 4              # Number of workers for data loading
  optimizer:
    name: Adam
    lr: 0.01
    weight_decay: 0.000003
  lr_scheduler:
    name: ReduceLROnPlateau
    mode: max
    factor: 0.5
    patience: 20
    verbose: true
  loss: BCEWithLogitsLoss
  metric: roc_auc_score
  num_epochs: 200             # Number of training epochs
  save_path: model.pth        # Path to save the model