Hi everyone,
I’m currently writing my bachelors thesis about the MA-POCA algorithm, and I was wondering how exactly the Neural Network is set up. I found that if you click on an .onx file in Unity, you can see all the layers used in the inspector. However, i do think this is only the network for the policy. Is there a way to figure out the network for the values and baselines? I would be eager to know if they use the same network as the policy and split later, or if they use totally seperate networks.
If anyone kowns where to find that information, it would be a great help!
Thanks a lot,
Leon 
Hi
The POCA trainer implementation is in the following folder of ML Agents:
ml-agents/ml-agents/mlagents/trainers/poca/
Hi,
thanks for the quick response. I do know where to find the MA-POCA implementation, however, I would like to know if there is already any form of visualization of the value and baseline neural network like this one for the policy that can be found in the inspector:
This shows me all 27 layers and all inputs in the Neural Network for my special project. Looking into the code of the MA-POCA algorithm sadly will not give me this information for the Baseline/Value. Is there any way to get this?
1 Like