shigeppp/deep-reinforcement-learning-algorithm-collection

Collection of Deep Reinforcement Learning Algorithms implemented in PyTorch.

Deep-Reinforcement-Learning

Collection of Deep Reinforcement Learning Algorithms in PyTorch.

Below a list of Jupyter Notebooks with implementations

Value Based / Offline Methods

Discrete Action Space

• Q-Learning     Source/Paper

• DQN      Paper

• Double DQN      Paper

• Dueling DQN      Paper

• N-Step DQN

• Noisy DQN      Paper

• Rainbow     Paper

Distributional RL

• Categorical DQN - C51     Paper

• QR-DQN

• IQN

• FQF

Continuous Action Space

• NAF - Normalized Advantage Function

-[Soft-DQN] TODO

---

Policy Based / Online Methods

Discrete Action Space

• Sarsa [Source/Paper]

• Vanilla Policy Gradient +LSTM [Source/Paper]

• A2C Paper

• A2C with gae* [TODO]

• A2C multi environment

• PPO Paper

• PPO with gae*

• PPO with gae and curiosity driven exploration (single, digit inputs) Paper

• PPO multi environment

Continuous Action Space

• A2C

• A2C with gae* [TODO]

• A2C multi environment

• PPO

• PPO with gae*PPO with gae multi

• PPO+curiosity&single+curiosity&multi

• PPO multi environment

gae* = Generalized Advanted Estimation Source

---

Actor-Critic Algorithms

• DDPG [Source/Paper]

• D4PG Source/Paper

• Twin Delayed DDPG (TD3)     Paper

• Soft Actor Critic (SAC-newest 2019 version)     Paper

---

Upside-Down-Reinforcement-Learning

Discrete and continuous action space implementation of ⅂ꓤ

---

Munchausen Reinforcement Learning

Implementierungen von Munchausen RL

• M-DQN

• M-IQN

• M-FQF

• M-SAC

---

Model-Based RL

---

Black-Box Optimization

• Evolution Strategies with mulit processing and novelty search

• Genetic Algorithm - GARNE

  • Genetic Algorithm implementation with LSTM, Multiprocessing over several CPUs and Novelty Search for Exploration

---

Multi-Agent Deep Reinforcement Learning

• Multi-Agent-DDPG

Hyperparameter Tuning

Gridsearch

Random Forest [TODO]

Genetic Algorithm [TODO]

====================================