UML Diagram

Relationships among classes and the organization of classes into namespaces can be viewed in the UML diagram linked here. Key things to note in this diagram are:

There five namespaces: gameboard, boardstate, piecepoints, moveslection, and jsonio
There are five anonymously namespaced classes: SpaceInfoProvider, PieceValueProvider, MoveEvaluator, JsonUtility, and BoardStateCoordinator.
Each of the anonymously namespaced classes is a CRTP interface and is implemented by one or two concrete classes from exactly one of the 5 namespaces.
The concrete-to-interface class pairings are:
- gameboard::GameBoard implements SpaceInfoProvider.
- boardstate::ZobristCoordinator implements BoardStateCoordinator.
- piecepoints::PiecePositionPoints implements PieceValueProvider.
- moveselection::MinimaxMoveEvaluator and moveselection::RandomMoveEvaluator each implement MoveEvaluator.
- jsonio::NlohmannJsonUtility implements JsonUtility.

Components

Each namespace, combined with the interface class it is paired with, can be considered a component of the AI engine. The primary responsibilities of each component are described below. In general, the class structure aims to prevent the engine's highest level concept - move selection - from coupling with implementation details of lower level components.

JsonIO

JsonUtilityInterface + jsonio namespace: Imports .json files to C++ objects such as a piecepoints::BPOPositionPoints. Currently, the sole implementation of the JsonUtility interface uses the Nlohmann json library. The interface keeps the rest of the codebase isolated from the details of the external library. If we ever want to use an additional json library and/or replace the Nlohmann, having the JsonInterface would simplify the integration process.

PiecePoints

PieceValueProvider + piecepoints namespace: Informs some implementations of a MoveEvaluator with the value associated with having a particular piece located at a particular gameboard::BoardSpace.

GameBoard

SpaceInfoProvider + gameboard namespace: Tracks the gameboard::BoardSpace locations of gameboard::GamePiece objects, and determines which moves of gameboard::GamePiece objects are allowed. Our concrete implementation of SpaceInforProvider, gameboard::GameBoard, accomplishes these tasks with the help of its gameboard::GameBoard. With the help of its gameboard::GameBoard.board_map_ and a gameboard::MoveCalculator.

Within the gameboard namespace, all standard rules of Xiangi are implemented, execept for some of the rules related to repeated moves. Specifically:

The perpetual checking and perpetual chasing rules are NOT implemented
Triple repeat move patterns with a periods of 2, 3, and 4 ARE prohibited (ABABAB, ABCABCABC, and ABCDABCDABCD are prohibited).
Triple repeat patterns with a period longer than 4 are NOT prohibited (ABCDEABCDEABCDE, and anything else of the form [X]ABCD[X]ABCD[X] is NOT prohibited).

BoardState

BoardStateCoordinator + boardstate namespace: Maintains an integer that represents the current board state, and updates the value of this board state integer whenever the gameboard::GamePiece configuration of a gameboard::GameBoard changes. Provides a MoveEvaluator with fast access to storage and retrieval of calculation results in a data structrue that usese board state integer values as keys.

boardstate::ZobristCoordinator, our concrete BoardStateCoordinator, supports using either a 32-bit, 64-bit, or 128-bit integer for its hashcalculator::HashCalculator.board_state_. The board state value is updated using a hashcalculator::ZobristKeys object combined with information from a gameboard::ExecutedMove.

MoveEvaluator

MoveEvaluator + moveselection namespace: Selects a gameboard::Move to execute. The moveselection::MinimaxMoveEvaluator implementation uses the recursive Minimax algorithm with the value of a particular board configuration determined based on piece- and position- dependent points obtained from a PieceValueProvider. This act of move selection represents the highest level concept on the C++ side of the project.

Previous	Next
Notation	Minimax Details