A tutorial on time series (temporal) train-test split

🎓 This tutorial aims to support beginners to forecasting learn how to perform a basic temporal train test split of a time series. By following the tutorial you will:

• ✅ Gain practical experience of how to perform a basic train test split of a univariate time series;
• ✅ Understand why you might consider a more advanced train-validation-test split;
• ✅ Build hands on experience of using both strategies to evaluate simple ARIMA models.
• ✅ Understand the limitations of basic train-test split strategies.
• 🎁 Bonus: Gain expertise in generating and plotting ARIMA prediction intervals.

License

The materials have been made available under an MIT license. The materials are as-is with no liability for the author. Please provide credit if you reuse the code in your own work.

Citation

If you reuse any of the code, or the tutorial helps you work, please provide a citation.

@software{TheOpenScienceNerd_TsSplit
author = {Monks, Thomas },
license = {MIT},
title = {{TheOpenScienceNerd - introduction to time series train test split}},
url = {https://github.com/TheOpenScienceNerd/temporal_train_test_split}
}

Installation instructions

Installing dependencies

All dependencies can be found in binder/environment.yml and are pulled from conda-forge. To run the code locally, we recommend installing miniforge;

miniforge is Free and Open Source Software (FOSS) alternative to Anaconda and miniconda that uses conda-forge as the default channel for packages. It installs both conda and mamba (a drop in replacement for conda) package managers. We recommend mamba for faster resolving of dependencies and installation of packages.

navigating your terminal (or cmd prompt) to the directory containing the repo and issuing the following command:

mamba env create -f binder/environment.yml

Activate the mamba environment using the following command:

mamba activate ts

Run Jupyter-lab

jupyter-lab

Repo overview

.
├── binder
│   └── environment.yml
├── CHANGELOG.md
├── CITATION.cff
├── LICENSE
├── ts_plotting.py
├── 01_ts_train_test_split.ipynb
└── README.md

• binder/environment.yml - contains the conda environment if you wish to work the models.
• 01_ts_train_test_split.ipynb - A tutorial notebook describing strategies for time series train test split
• ts_plotting.ipynb - a supporting module that contains plotly and other utility code.
• CHANGES.md - changelog with record of notable changes to project between versions.
• CITATION.cff - citation information for the package.
• LICENSE - details of the MIT permissive license of this work.