dl-gomoku's Introduction

Gomoku Game

Whoever align their five pieces first, vertically, horizontally or diagonally, wins the game.

How to play the game

It consists of a board with 11 * 11 (121) positions by default.
Each player, turn by turn, place their piece (black or white) on the available places.
The player, who places their pieces vertically, horizontally or diagonally first, wins the game.

A training data contains two states of the board: before making the move and after making the move.
- Before making the move means the current state of the board on which the player has to make the decision.
- After making the move means the next state of the board after the player has made the move.
Play the game yourself in which the opponent could be
- Random player - who places their pieces randomly on the board
- Myself - I play for both sides

The model detects the edges (points connecting where the pieces are placed) and generates the new edges as output.
Provide the training dataset where each data point is a tuple: (current_board_state, next_board_state).
- current_board_state: 11*11 matrix with values -1, 0 and 1.
  - 1 and -1 indicate the pieces placed by the two players.
  - 0 indicates the empty places.
- next_board_state: 11*11 matrix with values -1, 0 and 1 (as explained above).
The model, when provided with the current_board_state, should predict the next_board_state.

- Train the model by rewarding when the player is closer to winning and punishing when the player is closer to losing.

dl-gomoku's People

Implement the reinforcement learning strategy to win the game.

Existing

Use board search and apply the best strategy at each move (programmed logic)
Created a CNN model, trained on data created by the first implementation, and tested against other models.

Recommend Projects