Mava
Mava copied to clipboard
🦁 A research-friendly codebase for fast experimentation of multi-agent reinforcement learning in JAX
### Describe the bug It seems that the record_every flag is broken. It waits for an initial number of runs as specified by record_every before displaying an episode. But after...
### Describe the bug When running in single process mode, Mava tries to run everything on the GPU. This causes problems when we run multiple tests on a machine with...
## What? A recurrent IQL system, along with the necessary network, distribution and config files.
Adds mkdocs support for readthedocs. View the current feature branch docs here: https://id-mava.readthedocs.io/en/feat-mkdocs-documentation/ Additionally, the readme has been simplified as things like benchmarks have been moved to the docs pages....
## What? Fix and update the quickstart notebook based on the new changes in the repo. ## Why? The old code is outdated and we no longer use the `GlobalStateWrapper`...
## What? adds a wrapper for multicvrp. ## Why? To make mulctivcrp mava compatible. ## How? This wrapper takes in the output from the multicvrp environment and converts it into...
### Feature Create github action to automatically benchmark a PR to develop. We could run a simple rware test and grab the final reward. We'd need to have a minimum...
Currently the time limit, along with other `kwargs`, are set in the environment config and cannot be specified for individual scenarios. This is a problem as for some environments (e.g....
## What? Update the readme to mention that we now have support for continuous action space environments.