AI for Atari

Using two Deep Reinforcement Learning Algorithms to solve Atari 2600 Games respectively. First: implements a Double DQN with Prioritized Experience Replay (Proportional Prioritization). Second: implements Asynchronous Advantage Actor-Critic (A3C) algorithm.

Reference

The implementation of Double DQN with Prioritized Experience Replay (Proportional Prioritization) is based on:

• Mnih et al. Human-level Control through Deep Reinforcement Learning [2015.02]
• van Hasselt et al. Deep Reinforcement Learning with Double Q-learning [2015.12]
• Schaul et al. Prioritized Experience Replay [2016.02]

The implementation of Asynchronous Advantage Actor-Critic (A3C) algorithm is based on:

• Mnih et al. Asynchronous Methods for Deep Reinforcement Learning [2016.06]
• Babaeizadeh et al. Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU [2017.03]

Environment

• Python 2.7.x or 3.6.x
• NumPy 1.13.1
• TensorFlow 1.0.* or 1.1.* or 1.2.* or 1.3.*
• Keras 2.0.8
• SciPy 0.19.1 (For image pre-processing)
• H5py 2.7.1 (For saving or loading Keras model)
• Gym 0.9.3 (Provides Atari 2600 Games)