1
$\begingroup$

I'm trying to replicate the DeepMind DQN paper. I'm using OpenAI's Gym. I'm trying to get a decent score with Space Invaders (using SpaceInvaders-v4 environment). I checked the actions available with env.unwrapped.get_action_meanings(), and I get this:

['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE']

Checking the number of actions with env.action_space.n gives me a number of 6 actions.

The RIGHTFIRE and LEFTFIRE actions, I suppose, aren't used, given that they seem to do the same as LEFT and RIGHT, am I right?

If so, restricting the action size to the 4 first actions would improve my learning?

$\endgroup$
1
  • $\begingroup$ Programming questions (including how to use a certain library or API) are off-topic here. This type of question is more appropriate for Stack Overflow, even though the question is still related to Artificial Intelligence/Reinforcement Learning. Here, we focus on the theoretical, social, and philosophical aspects of AI. Please, see ai.stackexchange.com/help/on-topic for more details. $\endgroup$ – nbro Dec 28 '20 at 16:06
0
$\begingroup$

I found that we shouldn't eliminate those actions, because RIGHTFIRE and LEFTFIRE do RIGHT + FIRE and LEFT + FIRE, respectively. So they are different actions. I post this just to clarify in case someone faces the same doubt.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.