TicTacToe AI

A basic reinforcement learning AI with flexible tictactoe board size.

🎥 Brief example

🔨 Building

1. Clone the repository git clone https://github.com/Belluxx/TicTacToeAI.git
2. Go into the repository cd TicTacToeAI
3. Make a building directory mkdir build
4. Go into the build directory cd build
5. Run CMake cmake -DCMAKE_BUILD_TYPE=Release ..
6. Run make make

🕹️ Usage

First of all you need to choose an option:

[1] Load AI file to play against user
[2] Load AI file to play against another AI
[3] Benchmark an AI file
[4] Train new AI

If it is the first time that you run this project then you can either download a pre-trained AI file and choose option 1/2 or train a new one with option 4.

The option 3 is useful to check how strong your AI file is. It makes your AI play 100k games against a very weak agent that plays randomly and reports the result.

To train a new AI you will be asked for some tweaks:

1. Board size: the size of the tictactoe square; 3 means 3x3 square and 4 means 4x4 square.
2. Streak to win: The number of consecutive symbols needed to win, in the classic 3x3 square it is 3
3. Training iterations: how many times the AI will play against itself. Note that with a 4x4 board the iterations will take much more time than a 3x3 and more of them are needed to make it play well! For a 3x3 board i suggest 10 million iterations (It will take some minutes to complete) and for a 4x4 i suggest 200 million iterations (It will take some hours to complete)
4. File name: the name of the AI file that will be generated (if you put "test" the generated AIs for X and O will be respectively ai1_test ai2_test)

Then wait for the training to finish