Python project about Q-learning with approximate q-function

I am working on the Berkeley AI training project, I need to implement a search agent which chooses actions for the Pacman by using some reinforce learning method. the details for the project can be found on: [login to view URL] I try to use Q-learning with approximate Q-functions, but the performance of my code is really bad, my agent always repeat some actions while some other actions are never chosen, I can not figure out the problems, I need someone professional to help me with my code or built a new project. my code is in the '[login to view URL]' file in '[login to view URL]'