Welcome! This repo contains reinforcement learning algorithms.

• 2020.11.11 Confirmed for the performances of Value-based algorithms for several atari games such as Q-Bert, Breakout, Seaquest, Boxing, Pong, etc.

• 2021.02.15 Confirmed for the performances of Policy-based algorithms and distributed algorithms for CartPole and Lunarlander.

• I'll write up the detailed comments in all the codes soon.

• Colab online codes (Click 'colab' icon).

  • Value Based

    • DQN (1D Input) ,   DQN (2D Input)

    • DoubleDQN (1D input) ,   DoubleDQN (2D input)

    • DuelingDQN (1D input) ,  DuelingDQN (2D input)

    • NoisyNet (1D input) ,  NoisyNet (2D input)

    • Distributional (1D input) ,  Distributional (2D input)

  • Policy Based

    • REINFORCE in Discrete action space.

    • REINFORCE in Continuous action space.

    • REINFORCE with baseline in Discrete action space.

    • Actor-Critic in Discrete action space.

    • Actor-Critic in Continous action space.

• Ray Pararell python package Totorial series.

  • Ray Tutorial
  • Simple ReplayBuffer Tutorial
  • Simple ParameterServer Tutorial
  • Visualization with Ray Tutorial
  • Distributed DQN (with manipulating update steps)
  • Distributed DQN

• Paper References

  • DQN, Double DQN, Dueling DQN, PER, C51, Noisy DQN, Rainbow

  • REINFORCE, Actor-Critic - Sutton's Textbook

• Code References

  • Sutton - Reinforcement Learning Textbook 2nd ed.
  • https://github.com/Curt-Park/rainbow-is-all-you-need
  • https://github.com/MrSyee/pg-is-all-you-need
  • https://github.com/ShangtongZhang/DeepRL
  • https://github.com/sfujim
  • https://github.com/yandexdataschool/Practical_RL
  • https://github.com/seungeunrho/minimalRL