Full Game Environment Lag During Stage Transition about gym-super-mario-bros HOT 8 CLOSED

I've also opened #35 over at nes-py as the parallelism issue needs addressed to better support using multiple environments.

from gym-super-mario-bros.

This happens because a cutscene is playing the background on the NES emulator, but not being shown though gym because it's unnecessary noise. So, at the current moment this is intended behavior. However, I believe with RAM hacking of some form, the cutscene can be forced to end prematurely (as is done in several other points of this implementation). Will look into this.

from gym-super-mario-bros.

@Kautenja Thank you for reply!

from gym-super-mario-bros.

@Kautenja hi, it is another issue. I implement my RL algorithm with multiprocessing, but when some env reach End of World, other envs also stop.

from gym-super-mario-bros.

hmm how is the multiprocessing performed? They way the nes-py emulator works currently limits some of the parallelism approaches (one environment per Python process). However, the multiprocessing package when applied correctly should work fine.

from gym-super-mario-bros.

@Kautenja I think multiprocessing performed is not bad. i implement one env per one python process.
i will comment that issue on nes-py repo

from gym-super-mario-bros.

https://github.com/jcwleo/mario_rl
here is my repo :)

from gym-super-mario-bros.

Looks awesome! closing issue as parallelism is related to the parent code-base nes-py.

from gym-super-mario-bros.