TradeMaster is an open-source platform designed for quantitative trading (QT) using reinforcement learning (RL). It caters to researchers and practitioners by providing a comprehensive pipeline for developing, evaluating, and deploying RL-based trading strategies, covering data, simulation, algorithms, and evaluation.

How It Works

TradeMaster integrates six key modules: diverse financial market data, a full data preprocessing pipeline, high-fidelity market simulators for QT tasks, efficient implementations of over 13 novel RL algorithms, systematic evaluation toolkits with defined metrics, and interfaces for interdisciplinary users. This holistic approach aims to streamline the entire RL-for-QT workflow.

Quick Start & Requirements

• Installation: Installation tutorials are available for Linux, Windows, macOS, and Docker.
• Dependencies: Python 3.9 is specified. Specific dataset requirements are not detailed but links to data sources (Google Drive, Baidu Cloud) are provided.
• Resources: Colab versions are available for tutorials.
• Documentation: https://trademaster.readthedocs.io/en/latest/

Highlighted Details

• Supports over 13 novel RL-based trading algorithms, including DeepScalper, OPD, SARL, and PPO.
• Offers a Model Zoo with implementations in PyTorch and Ray.
• Includes a Visualization Toolkit with metrics like PRIDE-Star for systematic evaluation.
• Provides access to various datasets (S&P500, DJ30, BTC, SSE50, HK Stock) with different granularities.

Maintenance & Community

• Developed and maintained by the AMI group at Nanyang Technological University.
• Recent updates include new algorithms (EarnHFT, Market-GAN, MacroHFT) and features.
• Contact: TradeMaster.NTU@gmail.com

Licensing & Compatibility

• License: MIT.
• Compatibility: Suitable for commercial use and closed-source linking.

Limitations & Caveats

The platform focuses on RL-based strategies, potentially limiting applicability for non-RL quantitative trading approaches. Specific hardware requirements for training complex RL models are not detailed.