-
Notifications
You must be signed in to change notification settings - Fork 83
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
fix: shared weights with agent type #428
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks @AsadJeewa! 😄
Just see my few comments. Also, did we test this? At the very least we should do some runs on environments with and without different agent types.
I tested that everything works in multiple different environments with different (openspiel Tic Tac Toe, Petting Zoo Pong) or single-agent types (debugging, Petting Zoo Multiwalker, Flatland) to make sure that the networks were being assigned correctly and that execution proceeds. Since shared_weights default value is True, I think that we should run a small set of benchmarks @RuanJohn |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
- 1 potential issue in 3 places
- 1 debug print statement
- 1 minor comment
examples/tf/petting_zoo/atari/pong/recurrent/decentralised/run_madqn.py
Outdated
Show resolved
Hide resolved
This line (https://github.com/instadeepai/Mava/blob/develop/mava/specs.py#L68) assumes that all environments will follow the naming convention of |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
PR looks good to go from my side.
Thanks @AsadJeewa!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks @AsadJeewa! 👍
What?
Updated agent network types to only share weights between agent types
#427
Why?
When setting shared_weights to true, a single network is created for all agents. It should not be possible to share weights across agent types
How?
changed self._agent_net_keys in systems.py to assign the correct keys (for MADQN, MAPPO, MADDPG)
Extra
If we are happy with this change, would we then need to benchmark?