def __init__(self,memory_size=1000,memory_word_size=32):
self.Gmemory = nx.DiGraph()
self.StateAttributesDict={}
self.memory_size = memory_size
self.memory_word_size = memory_word_size
self.gamma = 0.9
self.lr = 0.01
self._write_content_similarity = utils.CosineSimilarity(1,self.memory_word_size,name="write_content_similarity")
self._TH = 0.99
def __init__(self,
num_actions=4,
num_node=10,
memory_size=100,
memory_word_size=100,
name='ControllerCore'):
#self.AbstractG = self._build_graph()
#由于目前还没有构造抽象图 20200430
#所以用原图代替
self.memory_size = memory_size
self.AbstractG = Memory.ExternalMemory(memory_size=self.memory_size)
self._read_content_similarity = utils.CosineSimilarity(
1, word_size=memory_word_size, name="read_content_similarity")
self.epsilon = 0.9
self.num_actions = num_actions
self.aggregator_cls = MeanAggregator
self.memory_word_size = memory_word_size
self.aggregator = self.aggregator_cls(self.memory_word_size,
self.memory_word_size,
name="aggregator",
concat=False)
def __init__(self,
num_actions=4,
num_node=10,
memory_size=100,
memory_word_size=32,
name='ControllerCore'):
# 外部存储的结构概览
self.memory_size = memory_size
self.memory_word_size = memory_word_size
# 构造抽象图,从Memory中重新实例化一个子图，作为抽象图
self.AbstractG = Memory.ExternalMemory(memory_size=self.memory_size)
self.aggregator_cls = MeanAggregator
self.aggregator = self.aggregator_cls(self.memory_word_size,
self.memory_word_size,
name="aggregator",
concat=False)
# 根据当前状态得到读取索引，用到相似度度量
self._read_content_similarity = utils.CosineSimilarity(
1, word_size=memory_word_size, name="read_content_similarity")
# 策略输出
self.epsilon = 0.9
self.num_actions = num_actions
self._optimizer = tf.keras.optimizers.Adam(learning_rate=1e-3)
def __init__(self, memory_word_size,memory_size,name='MemWriter'):
self._memory_word_size = memory_word_size
self._write_content_similarity = utils.CosineSimilarity(1,memory_word_size,name="write_content_similarity")
#应该把Gmemory继承过来做内部值，不然每个函数都要用一下
self._TH =0.999 #计算阈值，如果大于这个值新建节点(余弦距离的时候有正有负，所以暂时取正的)