TStarBot

A rule-based Star Craft II bot. Compatible with pysc2.agents.

Install

cd to the folder and run the command:

pip install -e .

Dependencies

pysc2 (Use Tencent AI Lab fork, required!)
pillow

We recommend pip install each Python package.

How to Run

Run the agent using the scripts from pysc2.bin. Example:

python -m pysc2.bin.agent \
    --map AbyssalReef \
    --feature_screen_size 64 \
    --agent tstarbot.agents.zerg_agent.ZergAgent \
    --agent_race zerg \
    --agent2 Bot \
    --agent2_race zerg

See more examples here.

Evaluate

Evaluate the agent (e.g., winning rate) using tstarbot.bin.eval_agent. Example:

python -m tstarbot.bin.eval_agent \
    --max_agent_episodes 5 \
    --map AbyssalReef \
    --norender \
    --agent1 tstarbot.agents.zerg_agent.ZergAgent \
    --screen_resolution 64 \
    --agent1_race Z \
    --agent2 Bot \
    --agent2_race Z \
    --difficulty 3

See more examples here. In particular, see how a well configured agent plays against difficulty-A (cheat_insane) builtin bot here.

Profiling

Use pysc2.lib.stopwatch to profile the code. As an example, see tstarbot/agents/micro_defeat_roaches_agent.py and run the following command:

python -m pysc2.bin.agent \
    --map DefeatRoaches \
    --feature_screen_size 64 \
    --max_episodes 2 \
    --agent tstarbot.agents.micro_defeat_roaches_agent.MicroDefeatRoachesAgent \
    --agent_race terran \
    --agent2 Bot \
    --agent2_race zerg \
    --profile

AI-vs-AI and Human-vs-AI

See examples here for AI-vs-AI and examples here for Human-vs-AI.

Coding Style

Be consistent with that of pysc2.

JohnyFeng/TStarBot2