Mava
Mava copied to clipboard
🦁 A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX
## What? - Add an environment wrapper that stacks new observation to past observations for training. ## Why? - This is one of the suggestion in Yu et al (2021)...
### Feature Terminate an experiment when one of the nodes has an error. This feature will be helpful in case of running multipple experiments, in case one of the experiment...
## What? Created a logger that can write data from the evaluator process to a structured json file. ## Why? This will enable for the usage of the EMARL repo...
## What? - Re-add obs network. - Remove assert that breaks on nested observations. ## Why? - Certain envs uses obs networks e.g. pong. ## How? - ## Extra -
### Please describe what needs to be maintained? Remove melting pot. Also remove Openspiel from setup.py. ### Describe the outcome you'd like Mava won't support melting-pot environment for the current...
Currently, we do not have tests for files in the IPPO folder. This includes the IPPO networks. It might be worthwhile including tests for some of these files.
### Please describe the purpose of the feature. Is it related to a problem? When using feedforward policies, partial observability can be handled by using multiple observations instead of only...
### Please describe the purpose of the feature. Is it related to a problem? It would be nice if we logged config/hyperparams on tensorboard. This will make doing things like...
### Please describe the purpose of the feature. Is it related to a problem? Arnu: The `on_execution_init_end` hook seems to play a similar role to the trainer's `on_training_utility_fns`, which creates...