Discussion: Parallel Gym Environments about retro HOT 6 CLOSED

I wrote a drop-in wrapper for retro to allow me to do exactly this, for use with the A3C algorithm in Tensorflow. In case unixpickle's answers don't work for you @mancoast, it's an option: retrowrapper. I just open-sourced it and stuck it in a repo.

from retro.

You can run multiple environments in parallel using a vectorized environment like SubprocVecEnv. This creates a separate sub-process for each environment. You can also create a gym.Env that samples a different environment on every reset(), like this JointEnv class, for example.

from retro.

I'm thinking of closing this issue since it's more of a discussion topic. Is there a better place to put this information?

from retro.

It is possible but only with multi-process communication. Unfortunately the emulation backend we're using cannot be loaded into the same process more than once. @unixpickle can tell you more about how to do the IPC.

from retro.

from retro.

@MaxStrange I've been trying to parralize an custom openai gym env that uses the airsim, but Things aren't working for me. The airsim has a physics engine in unreal and that needs to run in order to return the states which are images and collision information. I appreciate any advice that will put me on the right track.

from retro.