Pong-v0 takes forever to run with "AgentDQN" and crashes with "AgentActorCritic"

smilesun / rlR

Deep Reinforcement Learning in R (Deep Q Learning, Policy Gradient, Actor-Critic Method, etc)

https://smilesun.github.io/rlR

Other

26 stars 4 forks source link

Pong-v0 takes forever to run with "AgentDQN" and crashes with "AgentActorCritic" #22

Closed SebGGruber closed 6 years ago

SebGGruber commented 6 years ago

env = makeGymEnv("Pong-v0") agent = makeAgent("...", env) agent$updatePara(replay.mem = "UniformDB") perf = agent$learn(1) # I stopped here after waiting for over one hour

smilesun commented 6 years ago

I figured out it is storing process which slows things done, which is exactly the aim for the replayDB implementation you implemented? Have you test against the replayDB? It is easy to find out which one is the time consuming step. How fast is the storage of replayDB for example? @SebGruber1996

smilesun commented 6 years ago

@SebGruber1996 , I have removed the strange datatable in the replaymem by default, which greatly speed up the storage of image in pong.

smilesun commented 6 years ago

This problem does not exist anymore