/FinRL

A Deep Reinforcement Learning Framework for Automated Trading in Quantitative Finance. NeurIPS 2020. 🔥

Primary LanguageJupyter NotebookMIT LicenseMIT

FinRL: A Deep Reinforcement Learning Framework for Quantitative Finance twitter facebook google+ linkedin

Downloads Downloads Python 3.6 PyPI

News: we plan to share our codes for both paper trading and live trading. Please actively share your interests with our community.

FinRL is an open-source framework to help practitioners establish the development pipeline of trading strategies based on deep reinforcement learning (DRL). An agent learns by continuously interacting with an environment in a trial-and-error manner, making sequential decisions under uncertainty, and achieving a balance between exploration and exploitation. The open-source community AI4Finance (efficiently automating trading) provides resources about deep reinforcement learning in quantitative finance. It accelerates the paradigm shift from the conventional machine learning approach to RLOps in finance.

Join or discuss FinRL: AI4Finance mailing list, AI4Finance Slack channel:

Roadmaps of FinRL:

FinRL 1.0: entry-level for beginners, with a demonstrative and educational purpose.

FinRL 2.0: intermediate-level for full-stack developers and professionals. Check out ElegantRL.

FinRL 3.0: advanced-level for investment banks and hedge funds. Check out our cloud-native solution GPU-podracer.

FinRL 0.0: tens of training/testing/trading environments in NeoFinRL.

FinRL provides a unified DRL framework for various markets, SOTA DRL algorithms, benchmark finance tasks (portfolio allocation, cryptocurrency trading, high-frequency trading), live trading, etc.

Outline

Prior Arts

We published papers in FinTech at Google Scholar and now arrive at this project:

  • FinRL: A Deep Reinforcement Learning Library for Automated Stock Trading in Quantitative Finance, Deep RL Workshop, NeurIPS 2020.
  • Deep Reinforcement Learning for Automated Stock Trading: An Ensemble Strategy, paper and codes, ACM International Conference on AI in Finance, ICAIF 2020.
  • Multi-agent Reinforcement Learning for Liquidation Strategy Analysis, paper and codes. Workshop on Applications and Infrastructure for Multi-Agent Learning, ICML 2019.
  • Practical Deep Reinforcement Learning Approach for Stock Trading, paper and codes, Workshop on Challenges and Opportunities for AI in Financial Services, NeurIPS 2018.

Tutorials and News

Overview

A video about FinRL library. The AI4Finance Youtube Channel for quantative finance.

DRL Algorithms

We implemented Deep Q Learning (DQN), Double DQN, DDPG, A2C, SAC, PPO, TD3, GAE, MADDPG, etc. using PyTorch and OpenAI Gym.

Status Update

Version History [click to expand]
  • 2020-12-14 Upgraded to Pytorch with stable-baselines3; Remove tensorflow 1.0 at this moment, under development to support tensorflow 2.0
  • 2020-11-27 0.1: Beta version with tensorflow 1.5
  • 2021-08-25 0.3.1: pytorch version with a three-layer architecture, apps (financial tasks), drl_agents (drl algorithms), neo_finrl (gym env)

Installation

Installation (Recommend using cloud service - Google Colab or AWS EC2)

Clone the repository:

git clone https://github.com/AI4Finance-Foundation/FinRL.git

Install the unstable development version of FinRL using pip:

pip install git+https://github.com/AI4Finance-Foundation/FinRL.git

Prerequisites

For OpenAI Baselines, you'll need system packages CMake, OpenMPI and zlib. Those can be installed as follows:

Ubuntu

sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev libgl1-mesa-glx

Mac OS X

Installation of system packages on Mac requires Homebrew. With Homebrew installed, run the following:

brew install cmake openmpi

Windows 10

To install stable-baselines on Windows, please look at the documentation.

Create and Activate Python Virtual-Environment (optional but highly recommended)

cd into this repository:

cd FinRL

Under the root directory /FinRL, create a Python virtual-environment:

pip install virtualenv

Virtualenvs are essentially folders that have copies of python executable and all python packages.

Virtualenvs can also avoid packages conflicts.

Create a virtualenv venv under the root directory /FinRL

virtualenv -p python3 venv

To activate a virtualenv:

source venv/bin/activate

To activate a virtualenv on windows:

venv\Scripts\activate

Dependencies

The script has been tested running under Python >= 3.6.0, with the following packages installed:

pip install -r requirements.txt

Stable-Baselines3 using Pytorch

About Stable-Baselines 3

Stable-Baselines3 is a set of improved implementations of reinforcement learning algorithms in PyTorch. It is the next major version of Stable Baselines. If you have questions regarding Stable-baselines package, please refer to Stable-baselines3 installation guide. Install the Stable Baselines package using pip:

pip install stable-baselines3[extra]

A migration guide from SB2 to SB3 can be found in the documentation.

Stable-Baselines using Tensorflow 2.0

Still Under Development

Docker Installation

Option 1: Run a container via scripts

# grant access to execute scripting (read it, it's harmless)
$ sudo chmod -R 777 docker/bin

# build the container!
$ ./docker/bin/build_container.sh

# start notebook on port 8887!
$ ./docker/bin/start_notebook.sh

# proceed to party!

Option 2: Run a container manually

Build the container:

$ docker build -f docker/Dockerfile -t finrl docker/

Start the container:

$ docker run -it --rm -v ${PWD}:/home -p 8888:8888 finrl

Note: The default container run starts jupyter lab in the root directory, allowing you to run scripts, notebooks, etc.

Run

python main.py --mode=train

Backtesting

Use Quantopian's pyfolio package to do the backtesting.

Data

The stock data we use is pulled from Yahoo Finance API (the following time windows show the data split in the paper; users can customized to new time windows).

Contributions

  • FinRL is an open source library specifically designed and implemented for quant finance. Trading environments incorporating market frictions are used and provided.
  • Trading tasks accompanied by hands-on tutorials with built-in DRL agents are available in a beginner-friendly and reproducible fashion using Jupyter notebook. Customization of trading time steps is feasible.
  • FinRL has good scalability, with a broad range of fine-tuned state-of-the-art DRL algorithms. Adjusting the implementations to the rapid changing stock market is well supported.
  • Typical use cases are selected and used to establish a benchmark for the quantitative finance community. Standard backtesting and evaluation metrics are also provided for easy and effective performance evaluation.

Citing FinRL

@article{finrl2020,
    author  = {Liu, Xiao-Yang and Yang, Hongyang and Chen, Qian and Zhang, Runjia and Yang, Liuqing and Xiao, Bowen and Wang, Christina Dan},
    journal = {Deep RL Workshop, NeurIPS 2020},
    title   = {FinRL: A Deep Reinforcement Learning Library for Automated Stock Trading in Quantitative Finance},
    url     = {https://arxiv.org/pdf/2011.09607.pdf},
    year    = {2020}
}

Call for Contributions

Will maintain FinRL with the "AI4Finance" community and welcome your contributions!

Please check the Contributing Guidances.

Contributors

Thanks to all the people who contribute.

Support various markets

Support more markets for users to test their stategies.

SOTA DRL algorithms

Maintain a pool of SOTA DRL algorithms.

Benchmarks for typical trading tasks

To help quants have better evaluations, we will maintain benchmarks for many trading tasks, upon which you can improve for your own tasks.

Support live trading

Supporting live trading can close the simulation-reality gap, which allows quants to switch to the real market when they are confident with the results.

LICENSE

MIT License