This is a description of an amazing Deep Reinforcement Learning Suite

In recent years, we are constantly working on the implementation of various Deep Reinforcement Learning Methods, following the excellent research achievements from both industry and academic circles as shown in the diagram below.

We are applying ourself to developing an efficient, flexible and practical machine learning suite to share with Deep Learning researchers. Welcome to STAR, FORK or SHARE our code! More implementations of DRL Methods will be updated continuously!

• Policy Interation
  • Policy Evaluation
    • Policy Settings
    • Reward Achievement and getting Related Infomation
    • Feature Achievement from a new episode
    • Data Input
  • Policy Improvement
    • Generation of a new Policy
• Policy Base (comming for TensorFlow based version)
  • Action Selection
    • ε-Greedy
    • Probability Based
  • Policy Copy
• Policy Methods (waiting for more)
  • Deep Q-Learning
  • Double DQN
  • Monte-Carlo Q-Learning
  • Monte-Carlo Policy Gradient
  • Actor-Critic
  • Direct Policy Search (DPS)
• Data Generatior
  • Environment Base
    • Reset
    • Step
• Implementation of a Game
  • Atari Like
  • FrozenLake Like
• Model
  • Model base
    • Training
    • Testing
    • Parameter Achievement
    • Parameter Assignment
    • Model Restore
    • Model Save
  • Model Implementation (waiting for more)
    • MLP
    • CNN
    • Linear
    • GAN
    • AutoDecoder
• Configuration
  • Settings of hyper-parameters