[Solved] CS5225 Project 3-Deep Q-Learning

In this project, you will be asked to implement DQN to play Breakout. This project will be completed in Python 3 using Pytorch. The goal of your training is to get averaging reward in 100 episodes over 40 points in Breakout, with OpenAIs Atari wrapper & unclipped reward. For more details, please see the slides.

Deliverables

Please compress all the below files into a zipped file and submit the zip file (firstName_lastName_hw3.zip) to Canvas.

• Trained Model
  • Model file
  • If your model is too large for Canvas, upload it to a cloud space and write download.sh to download the model
• PDF Report
  • Set of Experiments Performed:
    • Include a section describing the set of experiments that you performed
    • what structures you experimented with (i.e., number of layers, number of neurons in each layer)
    • what hyperparameters you varied (e.g., number of epochs of training, batch size and any other parameter values, weight initialization schema, activation function)
    • what kind of loss function you used and what kind of optimizer you used.
  • Special skills: Include the skills which can improve the generation quality. Here are some tips may help. (Optional)
  • Visualization: Learning curve of DQN.
    • X-axis: number of time steps
    • Y-axis: average reward in last 30 episodes.
• Python Code
  • All the code you implemented including sample codes.

Hints

• Naive Pytorch Tutorial
• Official Pytorch Tutorial
• Official DQN Pytorch Tutorial
• Official DQN paper
• Rainbow: Combining Improvements in Deep Reinforcement Learning
• DQN Tutorial on Medium