Bank

Usage

MCTS

Run mcts_test.py to run the Monte-Carlo Tree Search algorithm against 7 other players in 1000 different games. The win/draw rates are reported in the console for each player.

Tournament Play

Run python tournament.py to run the tournament with the players configured in the tournament.py file.

Interactive Play

Running python interactive.py sets up an interactive game for a HumanPlayer against a few computer strategies.

Bank API Overview

This repo implements the Bank game logic as the foundation for the tournament ranking algorithm and Monte-Carlo tree search algorithm.

To simplify the player decision logic and Monte-Carlo tree search algorithm, the game is implemented as a turn-based game. After dice roll, each player is asked if they would like to bank. If a player chooses to bank, the decision process restarts, giving unbanked players a chance to change their mind based on the decision of others. The dice is rolled again once all players decide.

• BankState: Represents a single snapshot of a Bank game, containing all information needed to progress to the next state. This includes the current round number, the current player to decide, the pot amount, etc. The BankState objects should be treated as immutable.
• BankGame: Represents an entire Bank game, containing a history of previous states and events. Mutable, and can progress from state to state with the BankGame.decide method.
• BankPlayer: An abstract base class used to implement different Bank player strategies. The BankPlayer.get_decision method accepts a BankGame object, and makes a decision to bank or to pass given the current state of the game.
• HumanPlayer: An implementation of BankPlayer. Takes user input from the command-line. Useful for playing against computer strategies.
• RandomPlayer: Another implementation of BankPlayer. A basic strategy that randomly decides to bank with probability p regardless of the current game state, which is set in the constructor. Defaults to p=0.25.
• next_state: Returns the next state using the decision from the current player.
• play_game: Accepts a list of BankPlayers, and simulates a game with the players until completion. The game is returned. The event_hook argument allows the caller to handle state and event updates after each decision.