This project contains the custom code to train DMC agents on Keezen. The folder structure could be copied over a RLCard project. Just register the keezen environment in the __init__.py.

The folder structure:

/examples/experiments/dmc_keezen_result/keezenexpid7 -> experiment folder with a generated model and a processed.csv with tournament results against randomly playing agents.
/migrationtests -> Folder with game unit tests that were used during the migration from Swift to Python
/rlcard/envs/keezen.py -> the keezen environment
/rlcard/games/keezen/agent.py -> rule-based agent
/rlcard/games/keezen/board.py
/rlcard/games/keezen/card.py
/rlcard/games/keezen/cardop.py
/rlcard/games/keezen/game.py -> game implementation that uses other classes, like Board, Card, Move, Player, etc.
/rlcard/games/keezen/keezengameadapter.py -> adapter class to adapt the keezen game to the RLCard environment
/rlcard/games/keezen/move.py
/rlcard/games/keezen/player.py
/rlcard/games/keezen/rules.py
/tests/envs/test_keezen_env.py -> Test class for KeezenEnv
/tests/games/test_keezen_game.py -> Test class for KeezenGameAdapter

The project operated with the following scripts (in the examples directory):

create_graph_rnd.py -> Picks the last generated model and start a tournament against randomly playing agents.
create_graph_values.py -> Generates a graph with the results of tournament values.
evaluate_keezen_man.py -> Manual analyzer. Observe a game of four trained models.
evaluate_keezen_rb.py -> Run tournaments of trained models against rule-based agents.
evaluate_keezen_rnd -> Runs tournaments of trained models against randomly playing agents.
evaluate_keezen_versions.py -> Run tournament between two versions of trained models.
run_dmc_keezen.py ->Train DMC on Keezen.