deepwalk_flickr.yaml

application:
  graph

resource:
  gpus: []
  cpu_per_gpu: auto
  dim: 128

graph:
  file_name: <flickr.graph>
  as_undirected: true

build:
  optimizer:
    type: SGD
    lr: 0.025
    weight_decay: 0.005
  num_partition: auto
  num_negative: 1
  batch_size: 100000
  episode_size: 1000

train:
  # here the best setting uses no augmentation
  # in this case, DeepWalk is equal to LINE
  model: DeepWalk
  num_epoch: 2000
  negative_weight: 5
  augmentation_step: 1
  random_walk_length: 40
  random_walk_batch_size: 100
  log_frequency: 1000

evaluate:
  task: node classification
  file_name: <flickr.label>
  portions: [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]
  times: 5

save:
  file_name: deepwalk_flickr.pkl