search

krumiaa / MindMaker

MindMaker UE4 Machine Learning Toolkit
257 stars 31 forks source link

Training stopping short #15

Open mattgallivan opened 2 years ago

mattgallivan commented 2 years ago

I'm trying to run my training for 10k iterations (and that is set in "Trainepisodes" in the structure) but the training seems to stop short at ~391. Is there any reason it might quit early?

mattgallivan commented 2 years ago

It would appear this might be a SocketIO issue in U5 Preview 1. See this thread.

krumiaa commented 2 years ago

Still working on this, as mentioned seems to be SocketIO issue

FLOROID commented 2 years ago

In case this isn't already known - this also affects UE4

krumiaa commented 2 years ago

Should be resolved in the latest release, was an issue with python SocketIO libraries needing to be updated, will post a new requirements.txt or one can download the latest release from marketplace https://unrealengine.com/marketplace/en-US/product/neurostudio-self-learning-ai Please ensure the update works for you and close if so

FLOROID commented 2 years ago

So how would I best go about fixing this in an existing project?

FLOROID commented 2 years ago

I updated the MindMaker Plugin, however my Training still stops short. Do I have to create a new project and migrate my content over maybe?

FLOROID commented 2 years ago

Update: I used the MindMaker Plugin to create a new project and migrated my project files over. The issue still persists. However I did notice that the issue I was encountering only happened when using the PPO2 RL Algorithm. When using it the training would stop at exactly 394 when using 500 Training Episodes and 906 when using 1000 Training Episodes to give some examples. I went on to test all the different algorithms. Apart from DDPG which simply crashed the MindMaker executable and SAC which seems to require a different action space shape all the other algorithms worked as expected, producing actions up to the (for example) 500th Training Episode.

I hope this helps. If you need more information let me know.

krumiaa commented 2 years ago

Thanks for this feedback, will look into why some of these algorithms are working differently than others, its probably not something to do with the SOCKETIO changes since that would be affecting all algorithms equally. If you see any specific error codes when it stopping , let me know. Where you using Mindmaker with UE 4 or UE 5?

On Wed, May 11, 2022 at 12:31 PM Florian Dittrich @.***> wrote:

Update: I used the MindMaker Plugin to create a new project and migrated my project files over. The issue still persists. However I did notice that the issue I was encountering only happened when using the PPO2 RL Algorithm. When using it the training would stop at exactly 394 when using 500 Training Episodes and 906 when using 1000 Training Episodes to give some examples. I went on to test all the different algorithms. Apart from DDPG which simply crashed the MindMaker executable and SAC which seems to require a different action space shape all the other algorithms worked as expected, producing actions up to the (for example) 500th Training Episode.

I hope this helps. If you need more information let me know.

— Reply to this email directly, view it on GitHub https://github.com/krumiaa/MindMaker/issues/15#issuecomment-1123996119, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABI6N2JGIPZ4CNYDFTMTLADVJPOHRANCNFSM5PY6O4GA . You are receiving this because you commented.Message ID: @.***>