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


Checking the number of actions with env.action_space.n gives me a number of 6 actions. The 'RIGHTFIRE' and 'LEFTFIRE' actions I suppose isn't used, am I right? If so, restricting the action size to the 4 first actions would improve my learning?

  • $\begingroup$ It would learn faster since there are less actions to explore, but how did you conclude that RIGHTFIRE and LEFTFIRE actions aren't used ? $\endgroup$ – Brale Mar 13 '19 at 22:06
  • $\begingroup$ @Brale_ I tried and they do the same as LEFT and RIGHT. If I'm wrong please someone tell me. $\endgroup$ – JCP Mar 13 '19 at 22:52
  • $\begingroup$ Did you mean SpaceInvaders-v0? $\endgroup$ – Philip Raeisghasem Mar 14 '19 at 2:32
  • $\begingroup$ @PhilipRaeisghasem No, SpaceInvaders-v4. It's basically the same as v0 but improved. $\endgroup$ – JCP Mar 14 '19 at 2:49
  • $\begingroup$ Could you provide a link? I can't seem to find it $\endgroup$ – Philip Raeisghasem Mar 14 '19 at 2:50

Ok, I found that you shouldn't eliminate those actions cause RIGHTFIRE and LEFTFIRE do RIGHT + FIRE and LEFT + FIRE. So they are different actions. I post this just for clarify in case someone face the same doubt.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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