charliedream1/ai_quant_trade: a Chinese-language AI quant trading teaching repository

What ai_quant_trade actually is, and who it is written for

The repository describes itself as a one-stop AI quantitative trading platform covering learning, simulation and live trading. The structure tells a more specific story than the tagline. There is ai_notes for a knowledge base in Markdown and Jupyter, egs_trade for local strategy code, egs_alpha for factor libraries, egs_llm for large-model applications, egs_aide for auxiliary trading tools, egs_online_platform for hosted research platforms, and quant_brain as a core algorithm library. The README names three audiences: institutional investors, retail traders with programming background, and retail traders without one. The third group is the interesting claim. A repository of notebooks and per-directory READMEs can serve a non-programmer only if the examples run with minimal editing, and nothing in the README promises a guided path for that group beyond the directory-level documentation.

The primary language is Jupyter Notebook, so a large share of the value is in narrative cells rather than importable modules. If you want to read an explanation of why reinforcement learning is framed as goal-directed interaction rather than isolated prediction, that argument is present in the README and expanded in the notebooks. If you want a library you can pin and import, this is not it.

The mechanism: numbered example directories instead of a package

The README states plainly that the repository is not yet packaged as a Python module and instructs users to clone the whole project and enter each egs directory for its own usage notes. That single sentence determines the architecture. There is no top-level import surface, no versioned API, and no install target. What exists is a set of self-contained experiments, each with its own dependencies and its own README. The reinforcement learning section lists two entries: a prototype under egs_trade/rl/a001_proto_sb3 and a FinRL tutorial under egs_trade/rl/a002_finRL_tutorial/a01_Stock_NeurIPS2018, the latter tied to the paper Practical Deep Reinforcement Learning Approach for Stock Trading.

The README reports a backtest for that second entry on the Dow Jones 30: 53.1 percent annualized return, -10.4 percent maximum drawdown, Sharpe 2.17. Treat that number as a teaching artifact. It is a single reported backtest on a single index, with no walk-forward split, transaction cost model or slippage discussion in the README. The repository also exposes quant_brain and runtime directories at the top level, which suggests some shared code and a deployment path, but the README does not document their interfaces, so you have to read the source to know what they expose.

Installing ai_quant_trade and running a first example

There is no package on an index to install. The README gives a three-step clone, install, enter-directory sequence, and that is the whole setup story. The dependency file pins old versions, so expect friction on a current interpreter. The README lists Python 3.8 or higher in its badge, while requirements.txt pins python~=3.8.0 alongside pandas~=1.2.4 and numpy~=1.22.0. Create an isolated environment rather than installing into a system interpreter.

git clone https://github.com/charliedream1/ai_quant_trade.git
cd ai_quant_trade
pip install -r requirements.txt

After the install completes, the README directs you into a specific example directory to read its own documentation. The reinforcement learning tutorial is the one it names in the quick-start block.

cd egs_trade/rl/a002_finRL_tutorial/a01_Stock_NeurIPS2018

What you should see inside that directory is the material for the FinRL NeurIPS2018 walkthrough, not a command-line entry point. The README does not give a run command for it, so the next step is reading the files in that folder. If you would rather start with something that has a documented backtest framework attached, the README points at egs_trade/vanilla/double_ma, which it describes as a double moving average strategy plus a hand-written backtest framework with buy and sell markers plotted.

The Excel monitor is the most concrete tool in the repository

Most of the repository is educational. The exception is the V2 market monitor under egs_aide, which the update log dates to 2026.07.25 and describes as a modular excel_monitor package using a Sheet Handler pattern so each sheet refreshes independently. The feature list is unusually specific for this project: configurable price and percentage thresholds that turn a row red and raise a popup, in-Excel candlestick charts drawn with mplfinance, a sheet aggregating northbound capital flow with Weibo sentiment, news sentiment and stock forum activity, a searchable full-A-share universe with pinyin initial matching and a dropdown, and a multi-source fallback chain that lists qstock as primary with akshare, Eastmoney, Tencent, NetEase and efinance as backups.

Two details matter for anyone evaluating it. Configuration is hot-reloadable through YAML plus a 配置 sheet, so watchlist and refresh interval changes take effect without a restart. And the README claims pytest coverage of core logic across 156 tests. That is the only part of the repository making a stability claim backed by a test count, and it is also the only part framed as a daily-use tool rather than a lesson. The fallback design is a direct response to a real problem: Chinese retail data sources go down or change their endpoints, and a monitor that dies when one source fails is useless during a trading session.

Where the project stops short

Live trading is the weak point. The README's 实盘交易 section contains one entry, a Wind-based paper trading simulation of the double moving average strategy, and it states that Wind is typically the first choice of financial institutions and, because of its price, better suited to institutions. So the path from simulation to real orders runs through a paid institutional data terminal. There is no broker API integration documented in the README, no order state machine, no reconciliation, and no rollback procedure described anywhere in the documentation.

There is a second boundary. The project is Chinese-first. The README has an English version, but the directory names, the monitor's configuration sheet, and much of the documentation are in Chinese. An English-only reader will be navigating a repository where the file tree itself is partly unreadable. Third, the dependency pins are old enough that pandas 1.2.4 and numpy 1.22.0 will conflict with many current environments, and the README does not offer a container or lockfile alternative. If your team already runs a maintained backtesting framework, this repository is better read as a source of ideas and translated tutorials than adopted as infrastructure.

How it differs from Qlib, which the repository also covers

The README includes a directory called egs_trade/ms_qlib for Microsoft's Qlib framework, which makes the comparison internal to the project rather than external. The difference in approach is sharp. Qlib is a framework: it defines a data layer, a workflow configuration format and a model zoo, and you adopt its abstractions to get reproducibility. ai_quant_trade is a set of demonstrations that show you how to build or use those pieces, including a hand-written backtest framework in the double moving average example whose stated goal is understanding how a complete quant framework is constructed.

That means the two are not substitutes. If you need a pipeline you can run nightly across a universe of symbols, a framework gives you the scaffolding. If you need to understand why the scaffolding is shaped that way, or you want a worked Chinese-market example of factor mining with tsfresh or sentiment analysis with StructBERT, this repository is the more direct route. The practical consequence is maintenance: framework code is versioned and released, while example code in a teaching repository is updated when someone has time, which is why the update log here shows entries clustered in 2022, 2023, 2025 and 2026 rather than a steady release cadence.

Licence, maintenance and what upgrading costs you

The repository is Apache-2.0, which permits commercial use, modification and redistribution provided you keep the licence and notice files and state significant changes. That is permissive enough for internal enterprise use. Two things to check before relying on it. First, the example directories bundle or depend on third-party data sources and libraries, each under its own terms, and the Apache-2.0 grant on this repository does not extend to them. Second, the README promotes a WeChat public account and a paid knowledge community for video tutorials and additional model guides, and some content is described as living there rather than in the repository. That is a distribution choice, not a licensing problem, but it means the code you can read and the guidance you may need are in different places.

On maintenance: the last push was on 2026-09-06, and the repository is not archived. The only tagged release is v0.0.1 from 2025-02-15. There is no upgrade path to speak of because there is no package version to move between; upgrading means pulling the branch and re-reading whichever example directory changed. Given the pinned dependencies, the realistic cost of adopting this is the time spent rebuilding an environment that satisfies requirements.txt, not a recurring version-migration burden.

Editorial conclusion

Adopt this repository if you want worked examples in Chinese that link a paper to runnable code, particularly the FinRL NeurIPS2018 tutorial and the double moving average backtest. Do not adopt it if you need a pip-installable library, an English-first codebase, or a documented order-execution path: the README states the project is not packaged and points at Wind for live simulation, which it describes as suited to institutions. Before committing, verify what requirements.txt pins against your interpreter, and read the README inside the specific egs_trade subdirectory you intend to run.

Frequently asked questions

Official sources

1. charliedream1/ai_quant_trade on GitHub
2. Issues
3. License: Apache-2.0
4. README
5. Releases