UWaveGestureLibrary

Overview

Multivariate time series classification dataset from the UEA archive (Bagnall et al., 2018), built from the uWave gesture library (Liu et al., 2009). Each recording is one of eight simple gestures, given as the x, y and z acceleration of the motion (3 channels) over 315 steps.

../../_images/uwave_gesture_library_teaser.png

The loader passes the archive’s data through unchanged.

Train: 120 recordings (15 per class).
Test: 320 recordings (40 per class).

The archive labels the gestures 1.0 to 8.0; the class names are the gesture numbers 1 to 8 of the uWave paper, in that order.

Data Structure

Key	Type	Description
`series`	list of list of float32	Accelerometer recording, shape `(315, 3)`: time steps by channels
`label`	int	Gesture class (0-7)

Usage Example

import numpy as np
from stable_datasets.timeseries import UWaveGestureLibrary

train = UWaveGestureLibrary(split="train")
sample = train[0]
x = np.asarray(sample["series"])  # (315, 3)
print(x.shape, train.features["label"].names[sample["label"]])

References

Homepage: https://timeseriesclassification.com/description.php?Dataset=UWaveGestureLibrary
Archive record: https://doi.org/10.5281/zenodo.11206282
Original paper: https://doi.org/10.1109/PERCOM.2009.4912759
License: CC BY 4.0

Citation

@inproceedings{liu2009uwave,
  title={uWave: Accelerometer-based Personalized Gesture Recognition and its Applications},
  author={Liu, Jiayang and Wang, Zhen and Zhong, Lin and Wickramasuriya, Jehan and Vasudevan, Venu},
  booktitle={2009 IEEE International Conference on Pervasive Computing and Communications},
  pages={1--9},
  year={2009}
}

@article{bagnall2018uea,
  title={The UEA multivariate time series classification archive, 2018},
  author={Bagnall, Anthony and Dau, Hoang Anh and Lines, Jason and Flynn, Michael and Large, James and Bostrom, Aaron and Southam, Paul and Keogh, Eamonn},
  journal={arXiv preprint arXiv:1811.00075},
  year={2018}
}