// the find
bbfamily/abu
阿布量化交易系统(股票，期权，期货，比特币，机器学习) 基于python的开源量化交易，量化投资架构
Abu is a Python algorithmic trading framework targeting Chinese retail investors, covering A-shares, HK stocks, US stocks, futures, and crypto. It wraps backtesting, position sizing, ML-based strategy optimization, and a large library of technical analysis signals (Elliott Wave, Gann, harmonics, 100+ candlestick patterns) into a Jupyter-friendly interface. The target audience is Chinese quant hobbyists who want a batteries-included system without writing everything from scratch.
The UMP (umpire) system for filtering bad trades based on learned patterns from historical losses is a genuinely interesting idea — most retail frameworks skip this entirely. The modular factor architecture (buy/sell/position/stock-pick as separate composable classes) is clean and lets you mix strategies without rewriting glue code. The 30+ tutorial notebooks covering everything from basic backtesting to cross-validation and ML feature engineering are thorough and actually runnable. Grid search for hyperparameter optimization with custom scoring metrics is built-in, not bolted on.
The codebase is effectively frozen — last meaningful Python code commit was years ago, and the repo is now mostly a billboard for the commercial abuquant.com service. Data sources are tied to Chinese brokers and APIs that have broken or changed, so getting live data working requires significant effort. The project vendors its own copies of joblib, empyrical, and concurrent.futures rather than taking pip dependencies, which means you're stuck with old versions with no upgrade path. Documentation is entirely in Chinese with no English translation, making it inaccessible to most of the GitHub audience that might stumble on it.