Qlib is an AI-oriented quantitative investment platform designed for researchers and practitioners to explore, develop, and deploy quantitative investment strategies. It provides a comprehensive machine learning pipeline, from data processing and model training to backtesting and online serving, aiming to leverage AI for enhanced investment decision-making.

How It Works

Qlib employs a modular, loosely-coupled architecture, enabling individual components to be used independently. It features a robust learning framework supporting diverse paradigms like supervised learning and reinforcement learning, with specific models for market dynamics and continuous investment decisions. The platform integrates data processing, model training, backtesting, and online serving, facilitating end-to-end quantitative investment workflows.

Quick Start & Requirements

• Install: pip install pyqlib (for stable release) or install from source for the latest development version.
• Prerequisites: Python 3.8-3.12. Conda is recommended for environment management. Mac M1 users may need brew install libomp for LightGBM.
• Data: Official datasets are temporarily disabled; community-contributed data is available via wget or can be prepared using provided scripts.
• Docs: Qlib Documentation

Highlighted Details

• Supports a wide range of SOTA Quant research models, including GBDT, LSTM, Transformer, and RL-based approaches.
• Features an "Auto Quant Research Workflow" (qrun) for automated pipeline execution and graphical report analysis.
• Offers a modular interface for building customized research workflows by code.
• Includes solutions for adapting models to market dynamics (e.g., Rolling Retraining, DDG-DA) and RL for order execution.

Maintenance & Community

Qlib is actively developed by Microsoft. Contributions are welcomed, with guidance provided for new models, datasets, and documentation. Community interaction is encouraged via GitHub Issues and Gitter.

Licensing & Compatibility

The project is licensed under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The official dataset is temporarily unavailable. Some older models (e.g., TFT) have specific Python version dependencies (3.6-3.7) due to TensorFlow limitations. The run_all_model.py script currently only supports Linux and does not support parallel execution of the same model.