rosemarycox5334-debug/AlphaMaster

What it solves

AlphaMaster automates the process of discovering quantitative trading factors. It eliminates manual formula creation by using reinforcement learning to search through feature and operator spaces to find effective trading strategies based on historical market data.

How it works

The system operates through a unified pipeline for training, backtesting, and real-time analysis. It uses a reinforcement learning engine to generate factor formulas, which are saved as token sequences and executed via a StackVM. During backtesting and real-time execution, the system calculates a position size using the formula position = tanh(factor), ensuring that the logic used to evaluate historical performance is identical to the logic used for live market signals. The web interface allows users to manage data, monitor training progress, and receive real-time signal alerts.

Who it’s for

This project is designed for quantitative traders and researchers who want to automate the discovery of trading strategies using deep learning, specifically those who work with Parquet files or MetaTrader 5 (MT5) data and require a consistent environment for training and live signal monitoring.

Highlights

• Automated factor discovery using reinforcement learning.
• Unified logic for training, backtesting, and real-time signal generation.
• Integrated Web UI for managing the entire lifecycle of a strategy.
• Support for real-time signal monitoring with optional Feishu (Lark) notifications for trend reversals.
• Consistent backtesting metrics including Sharpe and Sortino ratios.