ffloof/turkishdraughts

A collection of ai, and guis for turkish draughts

turkishdraughts's Issues

Return results piece by piece

Instead of outputting the best board from the server, output boards and values as they come out of the engine.

Last major ui features

Arrows with values for best moves, (maybe with ai no move toggle).
Overall board position evaluation (sort of how lichess has in its analysis board)

Move check ordering

Optimize alpha beta pruning by checking promotion moves first.

Random R/W One Fail

The table is constantly failing exactly one read write test, not sure if its the last or the first. Either a problem with the table or the test.

Worker pool

Use a worker pool instead of the current parallel solution. Probably best to implement with transposition table to avoid having to communicate between threads constantly.

Use a transposition table

To optimize minmax use a transposition table to speed up searching duplicate board configurations by storing the results in a table.

Add Unit Tests and Benchmarking

Add unit tests for moving, capturing, and other board related functions.
Also implement benchmarks for minmax.

Improve heuristic of raw board value, and make alpha beta pruning more agressive

Same issue because alpha beta pruning got less effective when I updated raw board value.

Improve raw board value

Implement checks for pieces next to each other and overall position heuristics. I'd need to learn more about the game for this.

Start game and end game tables

Create a set of start moves and end moves to optimize end games and start games.

Swap Tile[] to a bitboard

Each tile can be represented by 3 bits, isEmpty, isKing, team. Using bits and bit shifting should make it faster to operate on the board and reduce memory accesses, increasing cpu usage.

Test edge case with takes

Taking and in the middle of a string of takes if theres a promotion, should it be able to promote and keep taking?

Cleanup take and move functions

Just particularly messy in general, but whole codebase needs to be commented/documented.

Tournament system

1. Add win condition to battles
2. Improve AI by banning previous boardstates to avoid permanent stalling
3. Implement logging of battles
4. New types of ai, dynamic minmax depth, random agent, one side minmax, one side minmax + regular minmax, and all the variants of existing ai
5. Implement tournament where each ai plays every other ai
6. Optionally add an elo system to the game
7. Maybe do a second third tournament with only the best ai and see how often they win on certain openings

Agressiveness

Encourage ai to trade down pieces if its equal, whether this be through heuristic or just in the move choosing order.

Look into using a modified verison of a chess hashing algorithm for transposition table

Minor optimizations

-Add a stack of previous moves in minmax to avoid repeat back and forth moves
-Optimize hashing algorithm, to use less memory overall by only hashing certain parts of board
-Dynamic depth based on board complexity(amount of initial moves) to keep move thought process time roughly the same in all stages of the game

Optimize king moves

Since ai rarely plays the king maybe reduce depth search of all king moves since a king can generate like 10+ moves each turn.