Deep-Q-Network--Breakout

This is the third assignment of ADLxMLDS course, NTU 2017 Fall, which is an implementation of Deep Q Learning (DQN) playing Breakout from OpenAI's gym with Keras.

Demo

I've implement deep reinforcement learning algorithm with Keras. In this project, the following techniques have been implemented:

• [x] Deep Q Learning
• [x] Double Q Learning
• [x] Dueling Network
• [ ] Prioritized Experience Play

Here's the architecture overview of DQN model in this work:

The project is derived from an assignment of the course ADLxMLDS 2017 fall.

Requirements

The follow packages are required, you can install them with pip3 install [package]

• opencv-python
• gym
• gym[atari]
• tensorflow
• keras
• scipy

Setup

• Training an agent to play atari game

  To train an agent playing Breakout with DQN, simply run

  python main.py --train_dqn

  To activate advanced algorithm such as Double DQN (DDQN)/ Dueling Network, you can run

  python main.py --train_dqn --ddqn True

  python main.py --train_dqn --dueling True

• Test the agent's ability

  Before test the code, download the pretrained model from google drive, and put the model into ./pretrained_model/model.h5

  By running the following command, you can get agent's average score in 100 episode

  python test.py --test_dqn

  Testing can be performed with the pretrained model training by default or with the model you trained by adding argument --test_dqn_model_path [your model path]

  To visualize the gaming progress, add --do_render to the end. You can also save it to vedio with --video_dir [path to save] (set smaller testing episode before doing so)

  python test.py --test_dqn --do_render

  or

  python test.py --test_dqn --video_dir [path]

References

• DQN in Keras + TensorFlow + OpenAI Gym by tokb23
• A tutorial of Prioritized Experience Replay by morvanzhou
• DQN-DDQN-on-Space-Invaders by yilundu
• Framework provided by TAs of the course.