AIFriendly-ticktacktoe
AI-Friendly plays tic-tac-toe
In a Nutshell
STEP 1: BUILD
bash build.shTo build the models yourself please run
- Building takes 5 minutes
- 2 information graphs are shown and script resumes after manually closing them
- Uses python3 and specifically Keras (pip install keras)
- Ends with a direct match between you and AIFriendly
STEP 2: PLAY
To play against AI-Friendly after building run
bash play-first.shIf you want to be the "X", or
bash play-second.shIf you want to be the "O"
Finally: if you want to make it more intelligent then run
bash enhace.sh(it can be enhaced as many times as you want, suggested: start with one)
If you want to play against the 'enhaced AI' then you must add the flag 'enhace' to the playing scripts e.g.
bash play-second.sh enhaceMore Info:
1) How did AI-Friendly learned to play Tic-Tac-Toe?
AI-Friendly was able to learn how to play Tic-Tac-Toe by finding patterns in a tic-tac-toe database.2) Does this problem differs from other problems that AI-Friendly solved?
Yes, the present task stands out in the context of it being what is called 'pattern recognition'3) What are the main differences between pattern-recognition tasks and pure-classification tasks?
In pure-classification tasks the objects tend to be uniquely defined: a chair is 100% a chair and a cat is 100% a cat; wether in this problem which involves pattern-recognition one same move in the training database may be sometimes tagged as "a good move" and other times as "a bad move".4) How was the database generated?
A training database was generated by defining a class with an attribute that makes a random move, instantiating two of such objects and finally making them both play against each other. The result was a collection of ~ 10k tic-tac-toe games5) Training with the randomly-generated database leads to an insatisfactory amount of intelligence. What can I do?
You can re-train it against itself and thus make it smarter by runningbash enhace.sh