My code is very poor in learning 2048 game using Double DQN

[codetiger]

Firstly, thanks for the great collection of code and articles. The articles were very useful in understanding DQN and implementing it.

However, my code is very bad in learning. I am not sure what is wrong with my code. I am using DDQN and passing rewards based on different criteria. Also the state is just a normalized version of the board itself.

My code repo is here https://github.com/codetiger/MachineLearning-2048
Let me know if you can review and help me understanding why my code doesnot learn anything even after 1000 episodes.

[dnddnjs]

Thank you for utilizing our code for your own code. We will review your code and tell you our opinion!

[codetiger]

Hi, I added some optimization techniques to my Agent and got better results.

The agent was trained for 100K episodes in 2x2 grid and got 100% optimal move every time. However, I did not have enough patience to train the agent for 4x4 grid. Updated my repo with new results