Mava
epymarl
Our great sponsors
Mava | epymarl | |
---|---|---|
5 | 3 | |
645 | 404 | |
5.7% | 6.7% | |
9.9 | 6.7 | |
3 days ago | 29 days ago | |
Python | Python | |
Apache License 2.0 | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
Mava
-
Starting wth Multi Agent Reinforcement Learning
If you want to play with models and algorithms around MARL, take a look at Mava.
-
Recomendations of framework/library for MARL
These are the main reasons we built Mava , a library built specifically for MARL. We are also in the process of rewriting it to be simpler to compose MARL components (communication, central training etc), and we are rewriting our codebase in JAX, so really looking forward to improved performance! (Disclaimer, I am one of the people working on Mava).
-
Training with multiple agents.
The OpenSpiel games (as well as several others) are also available from MAVA (https://github.com/instadeepai/Mava).
-
[R] Mava: a research framework for distributed multi-agent reinforcement learning
https://arxiv.org/abs/2107.01460 Mava, a scalable framework for research in multi-agent reinforcement learning, contains implementation for several multi-agent systems like multi-agent DQN (MADQN), MADDPG, MAPPO, MAD4PG, DIAL, QMIX, and VDN, and integrates well with multi-agent RL environments like PettingZoo, Flatland, OpenSpiel, RoboCup, and SMAC. On GitHub: https://github.com/instadeepai/Mava
epymarl
- EPyMARL with custom environment?
-
Recomendations of framework/library for MARL
If you're thinking of pettingzoo, you should try https://github.com/uoe-agents/epymarl. I use it actively for my work.
-
Simulation time for Multi-agent RL problems
Finally, since you're using PPO with centralised training with decentralised execution, you could perhaps also use its implementation in our group's MARL benchmark paper from NeurIPS 2021. I believe the PPO implementation from this benchmark was fast enough already.
What are some alternatives?
acme - A library of reinforcement learning components and agents
lingvo - Lingvo
tf2multiagentrl - Clean implementation of Multi-Agent Reinforcement Learning methods (MADDPG, MATD3, MASAC, MAD4PG) in TensorFlow 2.x
pymarl2 - Fine-tuned MARL algorithms on SMAC (100% win rates on most scenarios)
ai-economist - Foundation is a flexible, modular, and composable framework to model socio-economic behaviors and dynamics with both agents and governments. This framework can be used in conjunction with reinforcement learning to learn optimal economic policies, as done by the AI Economist (https://www.einstein.ai/the-ai-economist).
multi_agent_path_planning - Python implementation of a bunch of multi-robot path-planning algorithms.
PettingZoo - An API standard for multi-agent reinforcement learning environments, with popular reference environments and related utilities
agents-aea - A framework for autonomous economic agent (AEA) development
IC3Net - Code for ICLR 2019 paper: Learning when to Communicate at Scale in Multiagent Cooperative and Competitive Tasks
open_spiel - OpenSpiel is a collection of environments and algorithms for research in general reinforcement learning and search/planning in games.
maro - Multi-Agent Resource Optimization (MARO) platform is an instance of Reinforcement Learning as a Service (RaaS) for real-world resource optimization problems.