A Machine Learning Tic Tac Toe Game

[Barney]

From MiniMax to Machine Learning ... Tic Tac Toe is a good game for studying AI algorithm because it's simple!

I use Tabular Q Learning to implement this game, Every time a game finished, it will use the Q function to update

the score of each steps it played.

Q(S,A) = Q(S,A) + α ∗ (γ ∗ maxaQ(S′,a) − Q(S,A))

S being the current state, A the current action, S′ the state after doing A, α being the learning rate, γ being the

discount factor,  and maxaQ(S′,a) the highest Q value of any move in the next state S′, i.e. the Q value of the best

move in the following state.

It's funny to see that it plays better and better. That's why people were charmed by Machine Learning!

Thank you!

Download: tic_tac_toe.zip

[SamsonSlice]

Thank you very much for this example! This stuff blows my mind.

I couldn't get it to work for quite sometime and then I tried compiling it in 32 bit mode and it worked perfectly!