anubhavshrimal / reinforcement-learning

Implementation and Notes of different Reinforcement Learning Algorithms

Reinforcement-Learning

This repository contains the implementation and notes of different Reinforcement Learning Algorithms and techniques I have learned throughout the Udacity Deep Reinforcement Learning Nanodegree.

Each algorithm is used to solve different OpenAI gym environements.

Algorithms

1. Monte Carlo
2. Temporal Difference Algorithms
   • Sarsa(0)
   • Q-Learning / Sarsamax
     • CliffWalking-v0 environment
     • Taxi-v2 environment
   • Expected Sarsa
3. RL in Continuous Space
   • Discretization
   • Tile Coding
4. Deep Q-Network
   • Neural Network based
   • Convolutional Neural Network based
5. Double Deep-Q-Network
6. Dueling Deep-Q-Network

Projects

1. Navigation: Train an agent to collect yellow bananas while avoiding blue bananas using Deep Q-learning Algorithm.
2. Continuous Control: Train an robotic arm to reach target locations using DDPG Algorithm.
3. Collaboration and Competition: Train a pair of agents to play tennis using MADDPG Algorithm.

Dependency Installation

Follow the instructions given in the Installation_Guide.md to install the dependencies and run the code present in this repository locally.

RL Notes

I have also provided the notes I created while learning the above mentioned algorithms and techniques. You can find these notes in the RL-Notes folder.

1. The RL Framework
2. Monte Carlo Methods
3. Temporal Difference Methods
4. RL in Continuous Space
5. Deep Q Networks
   • Deep Q-Network (DQN)
   • Double DQN
   • Prioritized Experience Replay
   • Dueling DQN
   • Rainbow DQN

• Cheatsheet