初始化tf.train.Saver时“无值可保存”

Question

我有一个DDPG类，它是张量流神经网络。当我添加self.saver = tf.train.Saver()时会报告错误

  

“文件” C：\ Users \ Harry \ Anaconda3 \ lib \ site-packages \ tensorflow \ python \ training \ saver.py“，行1131，在_build中       引发ValueError（“没有要保存的变量”）“

这是我的代码的一部分。

 def __init__(self, action_dimension, state_dimension):
    self.memory = np.zeros((MEMORY_CAPACITY, state_dimension*2+action_dimension+1), dtype = np.float32)
    self.memory_pointer = 0
    self.sess = tf.Session()
    self.action_dimension = action_dimension
    self.state_dimension = state_dimension
    # define state space as x * state dimension matrix
    self.current_state = tf.placeholder(tf.float32, [None, state_dimension], 'current_state')
    self.next_state = tf.placeholder(tf.float32, [None, state_dimension], 'next_state')  # same as above
    self.reward = tf.placeholder(tf.float32, [None, 1], 'reward')
    self.saver = tf.train.Saver()

这是我在此类中声明的两个函数。

    def save_model(self):
    path = self.saver.save(self.sess, "/saved_model/model.ckpt")
    print("Model saved in path: %s" % path)

def load_model(self):
    self.saver.restore(self.sess, "/saved_model/model.ckpt")
    print("Model restored.")

我想知道我在做什么是错误的？谢谢。

Answer 1

可能是因为您尚未初始化变量。如果您在声明所有变量之后且在您之前运行tf global variables initializer self.saver你应该没事