a question in DDPG.train_on_batch

[lbgitjp]

I think below code maybe has problem:

PyTorch-DDPG-Stock-Trading/DDPG.py

Lines 85 to 90 in 0c5c2f9

	# train actor-critic by target loss
	self.actor_network.train(
	self.critic_network.train(
	y_batch, action_batch, state_batch
	)
	)

The two gradients need to be calculated separately, because their loss functions are different.
I think it should be changed to below:

`

    #for critic
    self.critic_network.train(y_batch, action_batch, state_batch)

    #for actor
    actor_loss =-self.critic_network.critic(self.actor_network.actor_action(state_batch), state_batch).mean()
    self.actor_network.optimizer.zero_grad()
    actor_loss.backward()
    self.actor_network.optimizer.step()

`
thanks!