K-Nearest Neighbors Time Series Prediction with Invariances (KNN-TSPI) algorithm implementation in python. For details about the model access the paper.

The package depends on the following third party libraries:

• numpy

Make sure you have pip package manager installed in your environment, then run:

pip install knn-tspi

Once the package is installed successfully, you can import the KNeighborsTSPI class and start forecasting univariate time series in a scikit-learn like manner.

import numpy as np

from knn_tspi import KNeighborsTSPI


data = 0.5 * np.arange(60) + np.random.randn(60)

model = KNeighborsTSPI()
model.fit(data)

y = model.predict(h=5)

For more detailed examples on how to use the package access examples.

Clone the repo with the command:

git clone https://github.com/GDalforno/KNN-TSPI.git

Create a python virtual environment and activate it with the command:

python3.11 -m venv .venv
source .venv/bin/activate

Then, set the enviroment up with the command:

make setup-dev

Once the setup is completed, launch the test suite for sanity check with:

make test-dev

You can contributing to the project by opening issues, pull requests and reviewing the code. It will help us a lot if you reference the package on blog posts, articlesm social media, etc.

This project is licensed under the MIT License - see the License file for details.

See the changelog for a history of notable changes to knn-tspi.

During my research in the field of application of machine learning to forecast time series in 2020, I stumbled with a lack of algorithms and frameworks specialized in this task.

One of my colleagues, Moisés Rocha, send me the paper of a modified KNN for time series prediction along with the experiment code written in MATLAB to help me out with my work. A coupled of days after it, I managed to port the code to both python and R and created this repo to store the resulted files.

Throughtout the years that followed, I have seen a growing interest in this repo, and now, I decided to publish it on pip to make it easier for people to include the model in their time series forecasting toolbox as I did a couple of years ago. As far as I know, there is no other implementation of the KNN-TSPI out there.

I am not planning on creating a CRAN package to distribute the model for the R community anytime soon. With that being said, feel free to implement it yourself if you wish. The core R code can be found here.

• Linkedin: https://www.linkedin.com/in/gabriel-dalforno-7b47ba227/

1. Parmezan, Antonio & Batista, Gustavo. (2015). A Study of the Use of Complexity Measures in the Similarity Search Process Adopted by kNN Algorithm for Time Series Prediction. 45-51. 10.1109/ICMLA.2015.217.