DDPG实现未收敛张量流

Question

我在训练TensorFlow模型时遇到问题，如果有人可以诊断出训练功能中的任何问题，我将不胜感激。它也可能是超参数或模型本身的问题，所以我已将代码放在GitHub

 def update(self, batch_size):
    states, actions, rewards, next_states, _ = self.memory.sample(batch_size)

    states = tf.convert_to_tensor(states)
    actions = tf.convert_to_tensor(actions)
    rewards = tf.convert_to_tensor(rewards)
    next_states = tf.convert_to_tensor(next_states)

    #Critic
    with tf.GradientTape() as tape:
        Qvals = self.critic([states, actions])
        next_actions = self.actor_target(next_states)
        next_Q = self.critic_target([next_states, next_actions])
        Qprime = rewards + self.gamma * next_Q
        critic_loss = tf.keras.losses.mean_squared_error(Qvals, Qprime)
    gradients_of_critic = tape.gradient(critic_loss, self.critic.trainable_variables)   
    self.critic_optimizer.apply_gradients(zip(gradients_of_critic, self.critic.trainable_variables))

    #Actor
    with tf.GradientTape() as tape:
        actions = self.actor(states)
        q_gradients = -self.critic([states, actions])
    parameter_gradients = tape.gradient(q_gradients, self.actor.trainable_variables)
    self.actor_optimizer.apply_gradients(zip(parameter_gradients, self.actor.trainable_variables))

    #Update target Critic
    target_param = self.critic_target.get_weights()
    param = self.critic.get_weights()
    for layer in range(0, len(target_param)):
        target_param[layer] = self.tau * param[layer] + (1.0 - self.tau) * target_param[layer]
    self.critic_target.set_weights(target_param)

    #update target Actor
    target_param = self.actor_target.get_weights()
    param = self.actor.get_weights()
    for layer in range(0, len(target_param)):
        target_param[layer] = self.tau * param[layer] + (1.0 - self.tau) * target_param[layer]
    self.actor_target.set_weights(target_param)

谢谢

https://github.com/kipgparker/Reinforcement