That is a program to train and test IA agent to play to a snake !
The algorithm used is the Q-learning.
This snake as some painful rules:
- He can only see a single line for each direction
- There is 2 Green apple making snake growth
- 1 Red apple making snake reduce
- Map border are closed
- chmod 777 install.sh
- ./install.sh
- wait for compile time
- ./learn2slither.sh
- have fun with your models !
'play' binary is here to show on defense that the playground effectively returns the good output, and proove the code is not cheating.
'l2s' is the CLI used for development. And is not fully implemented...
this wiki page is surprisely complete
- BellMan equation