This repository provides straightforward implementations of popular Deep Reinforcement Learning algorithms using TensorFlow 2. It targets students and researchers seeking to understand RL concepts, offering each algorithm within a single Python script for clarity and ease of study.

How It Works

The core approach emphasizes simplicity, with each algorithm encapsulated in a single Python file. Key techniques include using neural networks to approximate Q-functions (e.g., DQN), employing target networks for stability, and utilizing replay buffers for improved data efficiency. For partially observable environments, Recurrent Neural Networks (RNNs) like LSTMs are integrated. Algorithms like DDPG use deterministic actor models, while PPO incorporates importance sampling and clipping for stable policy updates.

Quick Start & Requirements

• Install: pip install tensorflow gym
• Run: python <Algorithm_Dir>/<Algorithm_Name>_<Action_Space>.py (e.g., python DQN/DQN_Discrete.py)
• Prerequisites: Python 3.x, TensorFlow 2.x, OpenAI Gym. GPU is recommended for performance.
• Docs: Spinning Up

Highlighted Details

• Implements DQN, DRQN, DoubleDQN, DuelingDQN, A2C, A3C, PPO, DDPG.
• Supports both discrete and continuous action spaces for applicable algorithms.
• Code structure prioritizes readability, with each algorithm in a single file.
• Includes core ideas and paper references for each implemented algorithm.

Maintenance & Community

The repository appears to be a personal project with no explicit mention of active maintenance, contributors, or community channels.

Licensing & Compatibility

The repository does not explicitly state a license. The code snippets use standard TensorFlow and Gym libraries, which are typically permissive.

Limitations & Caveats

TRPO, TD3, and SAC are listed as "Not yet implemented!". The project's personal nature suggests limited community support and potentially infrequent updates.