Tviskaron / pogema-baselines

PPO and PyMARL baseline for Pogema environment
20 stars 5 forks source link

New issue when running 32X32 and 64X64 #10

Closed 98luobo closed 2 weeks ago

98luobo commented 11 months ago

When I run 32X32 and 64X64, the program is killed by the system. I don't know if you have such a question?

Tviskaron commented 11 months ago

I haven’t experienced this specific issue. However, it might be related to insufficient RAM, especially if the problem does not occur with simpler configurations. I recommend monitoring your system’s memory usage during these runs.

Additionally, it’s worth noting that PyMARL hasn’t been updated for several years and I observed memory leaks on the part of my experiments...

98luobo commented 11 months ago

Yes, I have also observed memory bursting in my experiments. This code base is confusing to me...

98luobo commented 11 months ago

Hi, bor. I found a new way to work around this. Without pydantic, the program works better. Please wait for me to send my modified code to the comments section.

98luobo commented 11 months ago

New ISSUE: Hi, I was wondering how much memory do you actually need to run 32x32 and 64x64 experiments?

Tviskaron commented 11 months ago

Indeed, the PyMARL code is a bit outdated. In my opinion, it might not be worthwhile to delve deeply into identifying the specific error, as it's likely attributable to outdated dependencies that haven't been updated in a while.

I'm in the process of rewriting QMIX following the Clean RL paradigm. This update is expected to only address the memory leak issue.

Tviskaron commented 11 months ago

New ISSUE: Hi, I was wondering how much memory do you actually need to run 32x32 and 64x64 experiments?

I've tested it on a server with 256GB of RAM.