RL Programs train an agent to be able to learn to take actions that would maximize reward based on its interaction with an environment. In this project, the concept of Q-learning, a type of reinforcement learning is combined with deep neural networks to demonstrate deep-Q-learning.
-
main.ipynb: Here an agent tom has to learn to catch jerry in a grid environment that rewards tom as it gets closer to jerry in the shortest path possible. The report.pdf file describes this project in detail.
-
bonus.ipynb: Program to train an agent to play the ATARTI Breakout game using deep Q-networks and the stable baselines library. The bonus section in the report.pdf file shows the results of this program.
Note that small variations in results between the main.ipynb file and Report may be present due to random initializations and other randomness introduced during subsequent program runs.
Please refer Report.pdf for more details.