Skip to content

Latest commit

 

History

History
123 lines (97 loc) · 4 KB

README.md

File metadata and controls

123 lines (97 loc) · 4 KB
Raw

The Hive

Implementation of the board game "Hive" from gen42 [http://www.gen42.com/hive]. The aim of this project is to have a standard playground sandbox to develop and test A.I. to play the Hive game.

Board:

The board is a virtual concept since there is no board in the original board game but as pieces are placed next to each other an hexagonal structure forms. The internal representation of the hexagonal board is a 2 dimentional matrix.

  / \ / \ / \
 |0,0|1,0|2,0|
  \ / \ / \ / \
   |0,1|1,1|2,1|
  / \ / \ / \ /
 |0,2|1,2|2,2|
  \ / \ / \ /

Notation:

To represent each move of the game I created a human readable notation that can be easily parsed by a machine.

  • Piece: A piece is represented by 3 chars.

    • The first is the color and can be "black" or "white" {b|w}.
    • The second is the piece name {A|B|G|Q|S}.
    • The third is the piece number used to distinguish pieces of the same name {1|2|3}.
    • ex: wG2 - represents the second played grasshopper played by the whites.

  • Move: A move can be a placement of a piece from the player set into the board or moving a piece from its current position to a new position. Both of this actions are represented in the same way. A move representation has 3 sections: (action piece)+(point of contact)+(target piece)

    • The first section is the notation of the action piece.
    • The second is the point of contact with the piece where the move stops.
    • The third is the notation of the piece refered by the contact point.
    • ex: wG2/*bA1 - the second white grasshopper is positioned touching the first black ant on its left-upper side.

  • Point Of Contact: To represent how the pieces are placed next to each other we refer to wich face of the target piece the action piece will be touching by the end of the action.

    • /* - NW (north-west) active piece is place touching the target piece at its upper-left face.
    • |* - W (west) active piece is place touching the target piece at its left face.
    • \* - SW (south-west) active piece is place touching the target piece at its lower-left face.
    • *\ - NE (noth-east) active piece is place touching the target piece at its upper-right face.
    • *| - E (east) active piece is place touching the target piece at its right face.
    • */ - SE (south-east) active piece is place touching the target piece at its lower-right face.
    • =* - O (over) active piece is place on top of the target piece.

  • Starting Piece: The only exception to the previous notation is the starting piece of the first player. The first move of the first player is just the name of a piece since the placement of the 1st piece is always fixed.

  • Other Commands:

    • pass - Advances the turn counter and switches the active player. To be used when there is no possible move by a player. If both players pass then the game should end in a draw.

Development:

In order to run all the tests, you can use nose [https://nose.readthedocs.io/en/latest/testing.html]

Running the tests:

nosetests hivegame/test/

Running one particular test:

PYTHONPATH=. python hivegame/test/<test_name>.py

Running the game:

You can run arena.py for some game:

PYTHONPATH=. python hivegame/arena.py

Currently, you can change player types only in code. AI's can be chosen in the main function of arena.py. In a later version options will be provided in order to choose opponent.

Development of the AI

AI is based on AlphaGo Zero [https://deepmind.com/blog/alphago-zero-learning-scratch/] and is under development. You can test the training process of the current implementation by running:

PYTHONPATH=. python hivegame/train_test.py

Dependencies

The game is developed and tested with the package versions listed in dependencies.yml