config_seml.yaml

seml:
  name: pprgo
  executable: run_seml.py
  output_dir: '~/slurm_out'
  project_root_dir: .

slurm:
  experiments_per_job: 1
  sbatch_options:
    gres: 'gpu:"GeForce GTX 1080 Ti":1'
    mem: 32G                  # Main memory
    cpus-per-task: 5          # CPU cores
    time: 1-00:00             # Maximum runtime, D-HH:MM
    partition: [gpu_all]

fixed:
  data_dir: /nfs/shared/data/ # Directory containing .npz data files
  ntrain_div_classes: 20      # Number of training nodes divided by number of classes
  attr_normalization: None    # Attribute normalization. Not used in the paper
  topk: 32                    # Number of PPR neighbors for each node
  ppr_normalization: 'sym'    # Adjacency matrix normalization for weighting neighbors
  hidden_size: 32             # Size of the MLP's hidden layer
  nlayers: 2                  # Number of MLP layers
  weight_decay: 1e-4          # Weight decay used for training the MLP
  dropout: 0.1                # Dropout used for training
  lr: 5e-3                    # Learning rate
  max_epochs: 200             # Maximum number of epochs (exact number if no early stopping)
  batch_size: 512             # Batch size for training
  batch_mult_val: 4           # Multiplier for validation batch size
  eval_step: 1                # Accuracy is evaluated after every this number of steps
  run_val: False              # Evaluate accuracy on validation set during training
  early_stop: False           # Use early stopping
  patience: 50                # Patience for early stopping
  nprop_inference: 2          # Number of propagation steps during inference
  inf_fraction: 1.0           # Fraction of nodes for which local predictions are computed during inference

grid:
  split_seed:                 # Seed for splitting the dataset into train/val/test
    type: 'range'
    min: 0
    max: 5
    step: 1
  eps:                        # Stopping threshold for ACL's ApproximatePR
    type: choice
    options: [1e-2, 1e-4]

reddit:
  fixed:
    alpha: 0.5                # PPR teleport probability
    data_fname: 'reddit.npz'  # Name of .npz data file

remaining:
  fixed:
    alpha: 0.25               # PPR teleport probability
  grid:
    data_fname:               # Name of .npz data file
      type: choice
      options:
        - 'cora_full.npz'
        - 'pubmed.npz'
        - 'mag_large_filtered_06_09_coarse_standardized.npz'
        # - 'mag_large_filtered_06_09_fine_standardized.npz'