config.py

class NetworkConfig(object):
  scale = 100

  max_step = 1000 * scale
  
  initial_learning_rate = 0.0001
  learning_rate_decay_rate = 0.96
  learning_rate_decay_step = 5 * scale
  moving_average_decay = 0.9999
  entropy_weight = 0.1

  save_step = 10 * scale
  max_to_keep = 1000

  Conv2D_out = 128
  Dense_out = 128
  
  optimizer = 'RMSprop'
  #optimizer = 'Adam'
    
  logit_clipping = 10       #10 or 0, = 0 means logit clipping is disabled

class Config(NetworkConfig):
  version = 'TE_v2'

  project_name = 'CFR-RL'

  method = 'actor_critic'
  #method = 'pure_policy'
  
  model_type = 'Conv'

  topology_file = 'Abilene'
  traffic_file = 'TM'
  test_traffic_file = 'TM2'

  tm_history = 1

  max_moves = 10            #percentage

  # For pure policy
  baseline = 'avg'          #avg, best

def get_config(FLAGS):
  config = Config

  for k, v in FLAGS.__flags.items():
    if hasattr(config, k):
      setattr(config, k, v.value)

  return config