samuelemusiani/project_37635

A program made to play a generalized version of connect 4 perfectly

Connect X

A program made to play a generalized version of connect 4 perfectly

Commands

Compile

If you are on Linux or macOS you can simply run the following command in the project directory:

$ ./gradlew build

If you are so unlucky to work on Windows, the following command should work:

$ gradlew build

CXPlayerTester application:

After the project built successfully, you can test computer vs computer with the following command:

$ java -jar build/libs/connectx.jar 6 7 4 connectx.Players.L0 connectx.Players.L1  

If you are on Windows remember to change the / to the \ for the path.

There are additional flags to use in order to specify different configurations:

• With -v the output will be verbose
• With -t you can specify a maximum time of each move
• With -r you can specify how many rounds need to be played.

Verbose output and customized timeout (10 sec) and number of game repetitions (2 rounds) should look like this:

$ java -jar build/libs/connectx.jar 6 7 4 connectx.Players.L0 connectx.Players.L1 -v -t 10 -r 2

CXGame application:

In gradle.build the main class is specified, if you want another class to run as main (for example the CXGame) you can simply uncomment the commented line.

full_test script

There is a bash script with the name full_test.sh that allow to use the CXPlayerTester with all the board configurations stored in board_configurations.txt. The use is pretty simple: you just need to pass two player names in the command line and the script will do the rest.

./full_test.sh L0 L1

If you want the verbose output just use the -v flag.

project_37635's Issues

Aperture table

An aperture table is probably a good idea, especially in large board where we lost a lot of time searching for a move. The problem is that I don't now how to generate one or in general how to implement it.

Zobrist has slowed everithing

Before Zobrist on the board 6 7 4 we could reach depth 16 on the first move. I'm referring to the commit caccfe0.
After Zobrist (7303aac) we reach a depth 10 on the first move. Something wrong

Euristica

Serve una euristica veloce ma che abbia un senso

Killer-move heuristic

Need to read about it cause it can speed up a lot the search:

https://en.wikipedia.org/wiki/Killer_heuristic
https://www.chessprogramming.org/Killer_Heuristic
https://rustic-chess.org/search/ordering/killers.html
https://stackoverflow.com/questions/17692867/implementing-killer-heuristic-in-alpha-beta-search-in-chess

Lighter heuristic on small board

Nella tabella 4 6 4 gpt4reorder2 non vede abbastanza avanti per vincere come secondo giocatore. L9 però funziona, quindi ha senso alleggerire l'euristica per andare più avanti nella ricerca.

Forse la bitboard velocizza tutto :)

Report

In the end we need to write a small report on the project

BitBoard rappresentation for the small (<= 7x7) games

A faster way to represent the board in small games that can check if a move is winning in consta time. This should speed up the search. For large board we stick to normal CXBoard class

https://towardsdatascience.com/creating-the-perfect-connect-four-ai-bot-c165115557b0

HashMap vs Array

• Investigate while the HashMap hits are so low (20 - 100 top).
• Understand why with an array instead of the library HashMap the search is slower

License

Do we want to put a license on the project. If yes we need to verify that the code provided by the prof can be shared