A reinforcement learning model Q-learning used in simple maze game.
-
A training model on a simple maze game: blue square is the character; green square is the pickup which can increase, +1 point; then the red square is the trap that decrease, -1 point.
-
The model uses Q-learning, including main QValue Network and target QValue Network.
-
The final training on test may sometimes gets stucked, please try again before I fix it.I have fix it at 2017.12.03, and add another type of object, and a mode that randomly gives number of each kind of objexts.
- Choose to train or to test by setting the toTest and if you want to test, set testSteps to set the action steps that character allow to take.