Predict Backtest

This project uses PyTorch to train a neural network on patterns from preprocessed stock data. The trained network is then tested on new, unseen data to validate its accuracy. If the prediction is accurate, a backtesting library is used to simulate the results on future data.

Installation

Use Poetry, a package dependency manager, to install project dependencies:

poetry install

If you want to use your GPU, follow the instructions provided by your computer's GPU vendor to install the corresponding drivers, if not done so before.

Next, create a Jupyter kernel that points to the Poetry environment so you can work with project's installed dependencies in the Jupiter Notebooks in Jupyter later

poetry run python -m ipykernel install --user --name=<my_pytorch_env>

When you run the command to install the kernel (using ipykernel), it registers the kernel with Jupyter by creating a kernel specification file. These kernel spec files are typically stored in a user-specific directory. The exact location depends on your operating system:

On Ubuntu/Linux kernels are in: ~/.local/share/jupyter/kernels/

You can actually see all the kernels Jupyter knows about by running this command:

poetry run jupyter kernelspec list

Jupyter Notebook

If you don't have Jupyter Notebook installed in your system, in a separate folder, install it following:

python -m venv jupyter_env
source jupyter_env/bin/activate
pip install jupyter

Once jupiter environment activated and Jupyter installed open it with:

jupyter notebook

If you created a kernel with this project dependecies as explained in previous steps, you can select it on the top right hand side of the notebook.

Usage

Create a file config.json in the root folder of the project, and add your secret API key got the Tiingo API, which is used to download stock data.

{
  "tiingo-key": "your-api-key"
}

Open single_class.ipynb Jupyter Notebook and follow the instructions to train a neural network. You might need to adjust the download ticker and dates to get your required data. After executing the notebook, a file with the resulting neural network will be saved in the models folder. Also parameters about training results will be displayed.

The trained network obtained in the previous notebook can be used in single_class_test.ipynb to test on unseen data. Adjust the ticker and dates to test on the desired unseen data.

Notebooks and Git

Installed nbstripout to clean Jupiter Notebooks output before commiting changes to Git.

Executed command below to setup a Git filter to automatically remove the output cells (such as execution counts, outputs, and other metadata) from Jupyter Notebook files (.ipynb) before they are committed to a Git repository.

nbstripout --install --attributes .gitattributes

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

License

MIT