Skip to content

Latest commit

 

History

History
55 lines (32 loc) · 3.58 KB

Deep Reinforcement Learning.md

File metadata and controls

55 lines (32 loc) · 3.58 KB

Deep Reinforcement Learning in Games

In this project, you will implement a Deep Double Q Network that learns to play the popular Atari games. This was a major breakthrough in the field of Reinforcement Learning. The authors of the research paper could achieve superhuman performance on these games. The major reason why this was crucial is that the same agent (with separate training) could be used to play any of the Atari games. So, its time to explore this and flex your superhuman gaming skills!!

Expected Time required - 2-3 weeks

Pre-Requisites

  • Python

    This being a coding intensive project, one must have a background in Python. Skills in Numpy would be crucial, so brush up the basics of numpy before diving into the project.

  • Reinforcement Learning

    One must be familiar with the basics of Reinforcement Learning. One should know the discrete state algorithms like Q Learning etc. A slight introduction to the contiuous state space algorithms in Reinforcement Learning would be a plus. Experience in working with OpenAI Gym would be great.

  • Deep Learning

    One must be familiar with the basics of Neural Networks. It will be a plus if one has worked with deep learning frameworks like tensorflow or pytorch.

Note that these are not hard pre-requisites but being completely unaware of these topics would make it challenging to complete the project within 2-3 weeks.

Path to Accomplishment

  • Head here to brush up your Deep Learning knowledge. Read up the Week 4, 5, 6 materials to get a hang of how to use the deep learning frameworks.

  • Brush up the basics of Reinforcement Learning. These articles might be helpful -

  • Now, get started with Deep Reinforcement Learning especially Deep Q Networks. These articles might be helpful -

  • Get acquainted with OpenAI Gym.

  • Read this research paper to get into the details of the algorithm. This was a major breakthrough in the field of Reinforcement Learning.

  • Great! We know now all the theoretical aspects of the project. And its time to start coding.

  • Start coding up the algorithm as in the research paper. There are several developments in the field to boost the performance of the agent. So, once done with the basic algorithm as in the paper, one can look at the developments in this domain via various github repositories that have implemented this algorithm. One such repository is Atari Games

  • You can now train your agent locally or on Google Colab or any cloud service that you have access to. It might take a few days to train your agent to achieve superhuman performance but training for a few hours would be sufficient to obtain a decent performance.

  • Voila!! You are done. You can now flex your superhuman gaming skills.