This repository implements a simple rule-based agent and a DQN agent.
- Python 3.6
- Anaconda
- TensorFlow
- PySC2
- Baselines
First, you must install StarCraft II and create an account (it's free). You can refer to the following slides for installation: https://goo.gl/d5L4yD
Next, install the required packages. The following instructions are based on a macOS Sierra environment. When installing Anaconda, please press Enter and select "Yes" all the way through.
wget https://repo.continuum.io/archive/Anaconda3-5.0.0-MacOSX-x86_64.sh
bash Anaconda3-5.0.0-MacOSX-x86_64.sh
source .bash_profile
pip install tensorflow
pip install baselines
pip install pysc2
pip install absl-py
- scripted agent
python -m pysc2.bin.agent --map Simple64 --agent scripted_agent.simple_agent_protoss.RuleBaseAgent --agent_race protoss
- dqn agent
python dqn_agent/train_mineral_shards.py
- scripted agent
Consistently defeats the computer at the easiest difficulty level.
- dqn agent
Gets stuck around 13-14 points and cannot improve further.
Slides from a presentation at Taiwan R User Group / MLDM on 2017.10.02: https://goo.gl/oeEFvr
deepmind/pysc2 openai/baselines Building a Basic PySC2 Agent chris-chris/pysc2-examples xhujoy/pysc2-agents