DQN-tensorflow
TensorFlow implementation of Deep Q-Networks (DQN) for human-level control in reinforcement learning environments.
What is DQN-tensorflow?
DQN-tensorflow is an open-source implementation of Deep Q-Networks (DQN) using TensorFlow, based on the groundbreaking DeepMind paper 'Human-Level Control through Deep Reinforcement Learning'. It enables training reinforcement learning agents to play Atari games and other environments at human-level performance by combining Q-learning with deep neural networks. The implementation includes key stabilization techniques like experience replay and fixed target networks.
Machine learning researchers and practitioners interested in deep reinforcement learning, particularly those wanting to experiment with DQN algorithms in TensorFlow. It's also suitable for students learning about value-based reinforcement learning methods.
Provides a clear, well-structured TensorFlow implementation of the foundational DQN algorithm with essential stabilization techniques, making it easier to understand and experiment with deep reinforcement learning compared to building from scratch.
Overview
Tensorflow implementation of Human-Level Control through Deep Reinforcement Learning
Use Cases
Best For
- Implementing DQN algorithms for Atari game playing
- Learning deep reinforcement learning fundamentals with TensorFlow
- Experimenting with experience replay and target network techniques
- Comparing DQN variants like Dueling DQN and Double DQN
- Building reinforcement learning agents for OpenAI Gym environments
- Reproducing results from the original Human-Level Control paper
Not Ideal For
- Projects needing the latest reinforcement learning algorithms like PPO or SAC, as it focuses only on DQN variants.
- Teams using modern deep learning frameworks such as TensorFlow 2.x or PyTorch, due to its dependency on obsolete TensorFlow 0.12.0.
- Applications requiring continuous action spaces, since DQN is designed for discrete actions in environments like Atari games.
- Production systems demanding robust maintenance and scalability, given its experimental nature and lack of updates.
Pros & Cons
Pros
Provides a straightforward TensorFlow implementation of Deep Q-Networks, making it easy to grasp the core algorithm from the seminal paper, as shown in the well-documented code structure.
Includes experience replay memory and fixed target networks, critical for reducing correlations and stabilizing training, which are highlighted as key features in the README.
Offers implementations of Dueling DQN and Double DQN (DDQN), allowing for improved performance and experimentation with advanced DQN techniques, as evidenced in the results section.
Compatible with OpenAI Gym environments, enabling easy testing and training on a variety of reinforcement learning tasks, such as Atari games, with simple command-line usage.
Cons
Requires TensorFlow 0.12.0 and Python 2.7 or 3.3+, which are obsolete and may cause compatibility issues with modern systems, limiting its practicality for current projects.
The README admits that hyper parameters and gradient clipping are not implemented as in the original paper, which can affect the accuracy and reproducibility of results.
Focuses solely on DQN and its variants, excluding other reinforcement learning methods, making it less versatile for broader research beyond value-based approaches.
Frequently Asked Questions
Related Projects
KubeflowMachine Learning Toolkit for Kubernetes
Policy GradientDeep Learning and Reinforcement Learning Library for Scientists and Engineers
YOLO TensorFlow ++Translate darknet to tensorflow. Load trained weights, retrain/fine-tune using tensorflow, export constant graph def to mobile devices
Sentence Classification with CNNConvolutional Neural Network for Text Classification in Tensorflow
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.