Deep-Q Learning algorithm implementation using TensorFlow library

This project represents the implementation of the popular reinforcement learning algorithm called Deep-Q Learning introduced by Google Deep Mind. The reference to the paper is given on the following link:Human-Level Control through Deep Reinforcement Learning

The implementation consists of:

1. Deep Q-Learning
2. Using experience replay memory
3. Using two separate deep convolutional neural networks: Q and Target-Q network

Requirements

• Python 2.7 or Python 3.3+
• gym
• OpenCV2
• TensorFlow 0.11.0

Usage

To train and test a model for CartPole:

 $ python cartpole_dqn.py

To train and test a model for Pong:

$ python pong_dqn_conv.py

To train and test a model for Breakout:

$ python breakout_dqn_conv.py

Detailed results

[1] Accumulated reward, Q values and average error values obtained for CartPole environment:

[2] Accumulated reward, Q values and average error values obtained for Pong environment:

[3] Accumulated reward, Q values and average error values obtained for Breakout environment:

License

MIT licence.