VectorGym
Multi-process any(most) gym environment. Automatically parallel the given gym environment using multiprocessing; VectorGym forwards all properties and function (not starting with __) of the underlying gym to you.
Quick Start
Check this demo for a skeleton for training using VectorSim. It deals with only running unfinished environments during trajectory collection.
Run gym environment in parallel.
from VectorGym import VectorGym
if __name__ == '__main__':
envs = VectorGym('CartPole-v1', 2)
print(envs.action_space)
print(envs.observation_space)
envs.reset()
for _ in range(500):
envs.render()
actions = envs.action_space.sample()
res = envs.step(actions)
dones = [r[-2] for r in res]
envs.reset(select=dones)
envs.close()
Install
git clone git@github.com:MRzNone/VectorGym.git
cd VectorGym
pip install -e .