shenao-zhang / mra

Code for paper "Learning Meta Representation for Agents in Multi-Agent Reinforcement Learning".

Learning Meta Representations for Agents in Multi-Agent Reinforcement Learning

Code for paper "Learning Meta Representations for Agents in Multi-Agent Reinforcement Learning".

Requirements

To install requirements:

pip install -r requirements.txt

Training and Evaluation

To obtain the meta representation in 3 Markov Games in the resource occupying environment, run:

python main.py --use_gpu --env_id resource --model_name multi_3_scenarios

To train the MRA agents with different game settings, try to change the task_list and spec_list in main.py.

For example, change 6,9,12 to 3,6,12,24.

Acknowledgement

Some of our code is built upon the following repositories:

MAAC

EPC