Skip to content

A tic-tac-toe bot that you cannot defeat, written in JavaScript and based on a learning classifier system.

License

Notifications You must be signed in to change notification settings

azmke/lcs-tictactoe

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

15 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

lcs-tictactoe

lcs-tictactoe is a tic-tac-toe bot that you cannot defeat, written in JavaScript and based on a learning classifier system.

Screenshot

Let's play

In case you just want to play and find out what this tic-tac-toe bot is really capable of, we have something prepared for you. We trained the bot for a very long time and compiled the rule set into a single file that you can easily import.

  1. Open the index.html file in a browser of your choice.
  2. Under Import rule set click on Choose file and select the ruleset.json file.
  3. Click on Open file.
  4. Start playing!

Training

Feel free to train your own model with your parameter values.

Parameter Value Description
LEARNING_RATE 0 ... 1 To determine the new reward of a rule, the formula NEW_REWARD = (1 - LEARNING_RATE) * OLD_REWARD + LEARRNING_RATE * REWARD is used. A higher learning rate leads to a faster change of a rule's reward.
REWARD_DEFAULT n Sets the default reward for a rule that is added to the rule set for the first time.
REWARD_WIN n Sets the highest reward that can be included in the above formula (REWARD = REWARD_WIN) for a rule if the bot wins a game.
REWARD_TIE n Sets the highest reward that can be included in the above formula (REWARD = REWARD_TIE) for a rule if a game ends in a draw.
REWARD_LOSE n Sets the highest penalty that can be included in the above formula (REWARD = REWARD_LOSE) for a rule if the bot loses a game.
REWARD_DECREASE_RATE 0 ... 1 Only the bot's last move receives the highest reward/punishment. For each additional past move the reward/punishment is multiplied by (1 - REWARD_DECREASE_RATE). A higher decrease rate leads to a faster decrease of the reward/punishment for each move further back.
TRAINING_CYCLES 1 ... n Sets the number of games the bot will play during the training. A game is considered complete when the bot either wins, loses, or the game ends in a draw.

Statistics

The rule set coded in ruleset.json achieved the following performance during training:

Total training cycles 5,000,000 100.00 %
Total wins 4,373,340 87.47 %
Total ties 624,449 12.49 %
Total losses 2,211 0.04 %

A subsequent test run using the same rule set resulted in the following performance:

Total test cycles 10,000 100.00 %
Total wins 8,759 87.59 %
Total ties 1,241 12.41 %
Total losses 0 0.00 %

About

A tic-tac-toe bot that you cannot defeat, written in JavaScript and based on a learning classifier system.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published