setup

uv venv --python=3.10
source .venv/bin/activate
uv pip install -r requirements.txt
cd catanatron && uv pip install -r requirements.txt
cd catanatron/catanatron_core && uv pip install -e .
cd ../catanatron_experimental && uv pip install -e .
cd ../catanatron_gym && uv pip install -e .
cd ../.. && uv pip install -e .
python testing.py

strategy-game-agents

NEW WAY: Add to cli_players.py your agent (make sure to include init.py in directory) catanatron-play --players=LLM,R --num=1 --output=data/ --json

or

AZURE_OPENAI_API_KEY=azure_open_ai_key catanatron-play --players=LLM,R --num=1 --output=data/ --json

DEPRECIATED Example Use Case for Code catanatron-play --code=agents/vanillaLLM_player/vanillaLLM_player.py --players=AB,VanillaLLM --num=1 --output=data/ --json

How To View Commands catanatron-play --help

How To View Players catanatron-play --help-players

Current Players.

Code	Player	Description
R	RandomPlayer	Chooses actions at random.
W	WeightedRandomPlayer	Like RandomPlayer, but favors buying cities, settlements, and dev cards when possible.
VP	VictoryPointPlayer	Chooses randomly from actions that increase victory points immediately if possible, else at random.
G	GreedyPlayoutsPlayer	For each action, will play N random 'playouts'. Takes the action that led to best winning percent. First param is NUM_PLAYOUTS
M	MCTSPlayer	Decides according to the MCTS algorithm. First param is NUM_SIMULATIONS.
F	ValueFunctionPlayer	Chooses the action that leads to the most immediate reward, based on a hand-crafted value function.
AB	AlphaBetaPlayer	Implements alpha-beta algorithm. That is, looks ahead a couple levels deep evaluating leafs with hand-crafted value function. Params are DEPTH, PRUNNING
SAB	SameTurnAlphaBetaPlayer	AlphaBeta but searches only within turn
VLLM	VanillaLLMPlayer	Initial Vanilla LLM Player with no additions
LLM	LLMPlayer	LLM with adjusted prompt and code to fix bugs with vanilla llm
BL	BasicLangPlayer	First iteration of migration to using Lang Chain instead of custom base_llm class. Has inneficient memory that is saves number of messages which is set to a variable
TC	ToolCallLLMPlayer	LLM Player with tool call capabilities. Uses LangChain to call tools. Has access to web search, ...
FOO	FooPlayer	Player being created by creator agent