Skip to content

Latest commit

 

History

History
32 lines (29 loc) · 1.31 KB

README.md

File metadata and controls

32 lines (29 loc) · 1.31 KB

QMIX based on PARL and PaddlePaddle2.0

We reproduce the QMIX based on PARL and PaddlePaddle>=2.0.0, reaching the same level of indicators as the paper in StarCraft2 benchmarks.

QMIX

QMIX is a value-based multi-agent reinforcement learning algorithm.
Learn more about QMIX from: QMIX: Monotonic Value Function Factorisation for Deep Multi-Agent Reinforcement Learning

StarCraft2 Environment

Paper: The StarCraft Multi-Agent Challenge
Github Repositories: smac

Benchmark Results

Performance

  • We trained our model in 5 different scenarios: "3m", "8m", "2s_3z", "3s_5z" and "1c_3s_5z".
  • The difficulty in all scenarios are set to be "7" (very difficult).
  • We trained our model 3 times for each scenario.

How to Use

Dependencies

  • python3.6+

Start Training

  1. Modify the config in qmix_config.py.
  2. Start training:
    python train.py
  3. View the training process with tensorboard:
    tensorboard --logdir ./