RandomNet is a highly accurate and efficient time series clustering method. It has linear time complexity w.r.t the number of instances and the time series length. It not only achieves the SOTA performance but also performs well across all tested time series data types (device, ECG, EOG, EPG, image, motion, sensor, spectro, etc.).
This [paper] is accepted by Data Mining and Knowledge Discovery (DMKD)!
- Python 3.8
- NumPy
- TensorFlow 2.1
- PyMetis
- Scikit-learn
- Linux system
Install the required packages:
pip install -r requirements.txt
We use the UCR Time Series Classification Archive. You can download the full UCR datasets from [here].
To run the model on the Coffee dataset:
python RandomNet.py --dataset Coffee
@article{randomnet,
title={Randomnet: clustering time series using untrained deep neural networks},
author={Li, Xiaosheng and Xi, Wenjie and Lin, Jessica},
journal={Data Mining and Knowledge Discovery},
pages={1--30},
year={2024},
publisher={Springer}
}