tbgrl_cora.yaml

#
# Sample Weights & Biases YAML file for a hyperparameter sweep over all of TBGRL's parameters.
#
method: bayes
metric:
  goal: maximize
  name: val_mean_hits@50
parameters:
  base_model:
    distribution: constant
    value: triplet
  dataset:
    distribution: constant
    value: cora
  do_classification_eval:
    distribution: constant
    value: "false"
  drop_edge_p_1:
    distribution: q_uniform
    max: 0.9
    min: 0.1
    q: 0.1
  drop_edge_p_2:
    distribution: q_uniform
    max: 0.9
    min: 0.1
    q: 0.1
  drop_feat_p_1:
    distribution: q_uniform
    max: 0.9
    min: 0.1
    q: 0.1
  drop_feat_p_2:
    distribution: q_uniform
    max: 0.9
    min: 0.1
    q: 0.1
  epochs:
    distribution: constant
    value: 10000
  graph_transforms:
    distribution: constant
    value: standard
  intermediate_eval:
    distribution: constant
    value: "false"
  link_mlp_hidden_size:
    distribution: constant
    value: 256
  link_nn_epochs:
    distribution: constant
    value: 5000
  lr:
    distribution: q_uniform
    max: 0.1
    min: 0.001
    q: 0.001
  mm:
    distribution: q_uniform
    max: 1
    min: 0.9
    q: 0.01
  neg_lambda:
    distribution: q_uniform
    max: 0.95
    min: 0.05
    q: 0.01
  negative_transforms:
    distribution: categorical
    values:
      - scramble-edge-combo
      - rand-rand-combo
      - rand-rand-rand-combo
  num_runs:
    distribution: constant
    value: 5
  training_early_stop:
    distribution: categorical
    values:
      - "true"
      - "false"
  training_early_stop_patience:
    distribution: int_uniform
    max: 100
    min: 5
program: train_inductive.py