This project is an experiment about bringing reinforcement learing and Q-learning to a basic guessing game.
The game : A leader thinks of a secret number between 1 and 100 inclusive. The guesser then makes guesses at the secret number. After each guess, the leader tells the guesser if their guess was correct, in which case the game ends, too high, or too low.
In the project, a simple computed programm (see game.py) is the leader and an Q-learning agent is the guesser (see agent.py).
On a simulation like this, a perfect trained agent would be able to guess the number in 5.8 guess on average. In my experiment, it only succeed to reach 7.2.
- Clone the project :
git clone https://github.com/LouisonGitzinger/guessing_game.git
- Run
python main.py
http://personal.denison.edu/~kretchmar/pubs/WASP03.pdf
Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction. The MIT Press, 1998.