This repository contains the code for RATS - a rust-based rapid time-series data augmentation library and its python wrapper - RATSpy.
Rapid Augmentations for Time Series - RATS is a high-performance time-series data augmentation crate developed in Rust. RATS is available as a Rust crate and leverages Rust's core features as well as parallelism to speed-up the augmentation process. This library addresses the most common problem with respect to time-series data-augmentation: Performance bottlenecks. To enable cross-language integration, a python wrapper for RATS has been developed, namely, RATSpy.
📌 If you find this repository helpful or insightful, please consider giving it a star ⭐. If you use it in your research, please cite our paper (see Citation).
The popular python-based time-series library tsaug is used as a reference for comparing and benchmarking performance characteristics. On benchmarking against numerous datasets from the UCR archives, RATS and RATSpy are shown to be faster in performing the corresponding augmentations. At the same time, some augmentations are implemented which are not available in tsaug as added features.
The RATS crate provides many different augmenters. All augmenters are implemented in their own file in the augmenters module. They all implement the Augmenter trait which is implemented in src/augmenters/base.rs, allowing a common interface and the incorporation into an AugmentationPipeline which executes many arbitrary augmenters at once.
All augmenters work on the Dataset struct which holds a dataset or a batch of labeled univariate time series data.
The Rust crate can be found in the rats directory and the Python bindings in ratspy. All additional information regarding installation, usage and a detailed documentation of both these libraries can be found in the corresponding READMEs for rats and ratspy.
If you use this repository, adapt any of its components, or find it helpful for your work, please cite:
@misc{skaf-2026-RATS,
title={Rapid Augmentations for Time Series ({RATS}): A High-Performance Library for Time Series Augmentation},
author={Skaf, Wadie and Kern, Felix and Basu Roy, Aryamaan and Pradhan, Tejas and Kalkreuth, Roman and Hoos, Holger},
year={2026},
eprint={2601.03159},
archivePrefix={arXiv},
}If this repository is useful to you, a star ⭐ is also very much appreciated and helps others discover the project.
This work is licensed under the Apache License 2.0 - see the LICENSE file for details.