TuwTSG.v1 Benchmark for Time Series Clustering: Dataset Collection, Generator, and Evaluation Experiments

Please, if you use the TuwTSG.v1 benchmark or the TuwTSG tool in your research, cite our work as follows: @inproceedings{IglesiasZseby2026TuwTSG, author = {Iglesias, F{\'e}lix and Zseby, Tanja}, title = {On the Synthetic Evaluation of Time Series Clustering}, booktitle = {Proceedings of the 29th International Conference on Discovery Science (DS 2026)}, year = {2026}, address = {Mainz, Germany}, month = oct, note = {To appear}, } Technical details The repository is organized into five separate ZIP archives, which can be downloaded independently depending on the user's needs: TuwTSG.v1.zip — the TuwTSG.v1 synthetic dataset collection. generator.zip — the TuwTSG generator code, configuration files, and the HTML configuration wizard. evaluation.zip — the code, modules, and scripts used to run the evaluation experiments. docker.zip — Docker configuration files (Dockerfile, docker-compose.yml, Makefile, requirements.txt) to build the reproducible environment. docker_image.zip — a pre-built Docker image, provided to facilitate reproducibility without needing to rebuild the environment from scratch.

Users should download only the ZIP files relevant to their use case (e.g., only TuwTSG.v1.zip if just the dataset collection is needed) and extract them into the same local folder, so that the expected directory structure is preserved and all components work together consistently. The generator and evaluation code require Python 3.9.6 (Linux or macOS recommended), with dependencies listed in generator/requirements.txt and evaluation/requirements.txt .

Alternatively, the provided Docker image/configuration removes the need for a manual Python environment setup. Detailed step-by-step instructions for generation, evaluation, and Docker usage are provided in the accompanying README.md . Github repo The TuwTSG tool is also available in the Github repo: github.com/CN-TU/TuwTSG Further details The UCR datasets referenced in the evaluation pipeline were obtained from the TSB-Clustering and UCRArchiveFixes repositories.

While the TSB-Clustering repository's source code is MIT-licensed, this does not necessarily extend to the licensing of the underlying UCR datasets themselves; users should consult the original UCR Archive for the authoritative licensing and attribution terms. Full instructions for setup, dataset generation, environment configuration, and running the evaluation experiments (including expected folder structure) are documented in the README.md included with this artifact.