Deep Q-Learning Applied to Algorithmic Trading

The notebook with the experiments is available here The study is a replication and enhancement of the DQN by Theate, Thibaut and Ernst, Damien(2020).

Dataset

The data is provided by me in the data directory

GPU and Linux

Windows WSL

It's recommend to run this on a linux system. If you are on windows, install WSL on an admin powershell: wsl --install

Update your distribution: sudo apt-get update and sudo apt-get install wget ca-certificates. From your WSL cli, run code . and it will open a VSCode for you. Install the VSCode WSL extension for a seamless developer experience, see their tutorial here.

Cuda and GPUs

Install the NVidia drivers on your Linux box. Check if installation was good: nvidia-smi

Install TensorFlow's cuda framework: pip install tensorflow[and-cuda] and verify the installation: python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

Environment

Create the environment

conda create -n drl_ta python=3.11 && conda activate drl_ta && conda install -c conda-forge tensorflow-gpu cudatoolkit=11.8 cudnn yfinance matplotlib scipy && pip install --user tf-agents[reverb] ta tqdm pyarrow

or import the env from the JSON:

conda env create -f ./env.yml

if you get a TypeAlias error causing the Kernel to fail, upgrade it:

pip install --upgrade typing_extensions

Install FinRL

FinRL was one of the many inspirations. If you wish to try it, follow their installation guide.

sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-dev libgl1-mesa-glx
pip install git+https://github.com/AI4Finance-Foundation/FinRL.git

Try it out:

git clone https://github.com/AI4Finance-Foundation/FinRL.git
cd FinRL
python tutorials/FinRL_StockTrading_DQN.py

Credits and Citations

The DQN model was inspired by the paper:

@article{theate2021application,
  title={An application of deep reinforcement learning to algorithmic trading},
  author={Th{\'e}ate, Thibaut and Ernst, Damien},
  journal={Expert Systems with Applications},
  volume={173},
  pages={114632},
  year={2021},
  publisher={Elsevier}
}

adamd1985/Deep-Q-Learning-Applied-to-Algorithmic-Trading