parameter.py

FOLDER_NAME = 'backup'
model_path = f'model/{FOLDER_NAME}'
train_path = f'train/{FOLDER_NAME}'
gifs_path = f'gifs/{FOLDER_NAME}'
REPLAY_SIZE = 10000 # the size of replay buffer
MINIMUM_BUFFER_SIZE = 2000  # the minimum size of replay buffer before training
BATCH_SIZE = 64 
INPUT_DIM = 7   # the dimension of policy network input
EMBEDDING_DIM = 128 # the dimension of embedding vector
NODE_PADDING_SIZE = 360  # the number of nodes will be padded to this value
NORMALIZE_UTILITY = False # do you want to normalize the utility of nodes
K_SIZE = 20  # the number of neighboring nodes
USE_GPU = False  # do you want to collect training data using GPUs
USE_GPU_GLOBAL = True  # do you want to train the network using GPUs
NUM_GPU = 1 # the number of GPUs you want to use
NUM_META_AGENT = 24 # the number of meta agents
LR = 1e-5   # learning rate
GAMMA = 1   # discount factor
DECAY_STEP = 256  # not use
SUMMARY_WINDOW = 50
LOAD_MODEL = False  # do you want to load the model trained before
SAVE_IMG_GAP = 500  # save image every SAVE_IMG_GAP episodes