This toolkit provides researchers and practitioners with a framework for recording and replaying cryptocurrency limit order book data, enabling the training of Deep Q-Network (DQN) agents for algorithmic trading strategies. It focuses on data handling and agent training, offering a research-oriented environment without live trading capabilities.

How It Works

The project utilizes a modular design to record full limit order book and trade tick data from exchanges like Coinbase Pro and Bitfinex, storing it in an Arctic Tickstore (MongoDB). It then replays this historical data to derive feature sets, which are fed into a DQN agent implemented using Keras-RL. The gym_trading module provides an extended OpenAI Gym environment for observing order book data, and technical indicators are implemented for O(1) time complexity.

Quick Start & Requirements

• Install: pip3 install -e . (after setting up a virtual environment and installing dependencies).
• Prerequisites: Python 3.x, MongoDB, TensorFlow/Keras (specific versions listed in README), Keras-RL, Arctic library (pip install git+https://github.com/manahl/arctic.git).
• Setup: Requires manual installation of specific TensorFlow and Keras-RL versions. Data recording and feature extraction are separate steps.
• Links: GitHub Repository

Highlighted Details

• Records limit order book and trade tick data from Coinbase Pro and Bitfinex.
• Implements stationary technical indicators with O(1) time complexity.
• Provides an extended OpenAI Gym environment for trading simulations.
• Uses a DQN agent with Keras-RL for training.

Maintenance & Community

The project was last updated in September 2021 with dependency updates. The "FULL" branch is noted as the foundation for future development, while other branches are marked as unmaintained since October 2018.

Licensing & Compatibility

The repository does not explicitly state a license in the README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is strictly for research purposes and does not support live trading. Specific TensorFlow and Keras-RL versions are required, which may conflict with other RL platforms. Several branches are marked as unmaintained.