Records of Reinfocement Learning Experiments

Records of Reinforcement Learning Experiments

Background and Reference course:

I have been following the Movan’s RL course for more than a month.
At first, I found some of concepts really hard to understand, then I started programming these algorithms, I perhaps comprehended the meaning of these concepts and the principle of agent updating.
So thanks to Movan and I also want to share my experiments for others.
As for why to use English, because of the request of my boss.
It’s always right to practice a little more at the normal times.

the natural DQN with the env of Maze

the codes has been uploaded to my github —— thank your stars~

In this codes, you can customize the environment, such as resetting maze size, resolution and penalty and reward points.
as you can see, the maze is bigger than Movan~ when you train your agent to learning how to get reward with simple neural network will be so difficult.

my natural DQN is same as Movan’s, the tensor graph is ——