contact: sezan92[at]gmail[dot]com
In this project I have tried Reinforcement Learning on two Open AI games . One is Mountain Car Game and another is AirRaid, an Atari Game.
For the Mountain Car game , I have used Q Learning ALgorithm . The Agent needs to earn -110 reward . We have achieved the reward by running upto 30000 episodes. For the simple Q learning algorithm, I have taken help from this blog . The Video of Full training is here . Please notice that the video is full training , so it is very large . Skip to last 1-2 minutes to watch the final Result . The Code MountainCarQLearning.py
is the Code for Version 1.
After almost 6 months, I have noticed some bugs in my original code. I have took help from Max's Blog . His implementation of Q learning is better but quite difficult to understand. I have revised the Q learning Class. My implementation is only 34 lines compared to 91 lines of his code! The file MountainCarQLearning2.py
is the Updated and better version. The file MountainCarQLearning-v2.ipynb
is the Jupyter notebook of the same code. The file QLearn.py
is the Q learning Implementation Class. The updated version's video is here
For the AirRaid Game, I have used Deep Q Network with Experience replay . For the Architecture I have used the architecture mentioned here .