wandb_sweep_kge.yaml

program: train.sh
# If you only want to run all combinations of parameters once, set method: grid.
method: bayes
name: biomedgps-kge
metric:
  goal: maximize
  name: Test_HITS@10
parameters:
  # All data files must be in the data_path/${dataset}. The data files must be named as train.tsv, valid.tsv, test.tsv.
  data_path:
    values:
      - "/data/biomedgps-data/wandb/data"
    distribution: categorical
  dataset:
    values:
      - "drkg+hsdn+custom+malacards"
    distribution: categorical
  model_path:
    values:
      - "/data/biomedgps-data/wandb/models"
    distribution: categorical
  wandb_entity:
    values:
      - "yjcyxky"
    distribution: categorical
  wandb_project:
    values:
      - "biomedgps-kge-auto"
    distribution: categorical
  enable_embedding:
    values:
      - False
    distribution: categorical
  # If you enable embedding, please ensure that the related embedding files are in the data_path/${dataset}. These embedding files must match with the hidden_dim. such as entities_embeddings_400.tsv, relation_types_embeddings_400.tsv
  hidden_dim:
    values:
      - 400
      - 768
      - 1024
    distribution: categorical
  lr:
    max: 0.2
    min: 0.05
    distribution: uniform
  batch_size:
    values:
      - 2048
    distribution: categorical
  model_name:
    values:
      - "ComplEx"
      - "TransE"
      - "DistMult"
    distribution: categorical
  gpu:
    values:
      - 0
    distribution: categorical
  num_proc:
    values:
      - 4
    distribution: categorical
  neg_sample_size:
    values:
      - 256
    distribution: categorical
  max_step:
    values:
      - 100000
    distribution: categorical
  log_interval:
    values:
      - 1000
    distribution: categorical
  batch_size_eval:
    values:
      - 16
    distribution: categorical
  regularization_coef:
    values:
      - 1.00E-07
    distribution: categorical
  neg_sample_size_eval:
    values:
      - 10000
    distribution: categorical
command:
  - bash
  - train.sh
  - ${args}