actor_network.py 文件源码

def __init__(self,sess,state_dim,action_dim,scope):

        self.state_dim = state_dim
        self.action_dim = action_dim
        # create actor network
        self.state_input,self.action_output,self.net = self.create_network(state_dim,action_dim,scope)

        # create target actor network
        self.target_state_input,self.target_action_output,self.target_update,self.target_net = self.create_target_network(state_dim,action_dim,self.net,scope)
        # define training rules
        if scope != 'global/actor':
        self.q_gradient_input = tf.placeholder("float",[None,self.action_dim])
        self.parameters_gradients = tf.gradients(self.action_output,self.net,-self.q_gradient_input)
        global_vars_actor = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES, 'global/actor')
        self.optimizer = tf.train.AdamOptimizer(LEARNING_RATE).apply_gradients(zip(self.parameters_gradients,global_vars_actor))
    sess.run(tf.global_variables_initializer())

        #self.update_target()
        #self.load_network()