HandMovementDirection

Task: Classification Classes: 4 Channels: 10 Length: 400

Overview

Multivariate time series classification dataset from the UEA archive (Bagnall et al., 2018), based on data set 3 of BCI Competition IV (Tangermann et al., 2012), recorded by Waldert et al. (2008). Two healthy subjects moved a joystick with the right hand and wrist from a center position toward one of four self-chosen targets while their brain activity was recorded with MEG. Each trial covers 0.4 s before to 0.6 s after movement onset: 10 MEG channels above the motor areas, resampled to 400 Hz (400 steps).

../../_images/hand_movement_direction_teaser.png

The archive keeps the competition’s train/test split but merges the two subjects. The test set is small, so one trial changes test accuracy by 1.35 percentage points, and the task is hard: the published Multiverse results on this split range from below the majority-class rate (1NN-DTW) to about 51 % (ROCKET). The loader passes the archive’s data through unchanged.

  • Train: 160 trials (40 per class).

  • Test: 74 trials.

Class names are the archive’s labels in lower case: backward, forward, left, right.

Data Structure

Key

Type

Description

series

list of list of float32

MEG recording, shape (400, 10): time steps by channels

label

int

Movement direction (0-3)

Usage Example

import numpy as np
from stable_datasets.timeseries import HandMovementDirection

train = HandMovementDirection(split="train")
sample = train[0]
x = np.asarray(sample["series"])  # (400, 10)
print(x.shape, train.features["label"].names[sample["label"]])

References

Citation

@article{tangermann2012review,
  title={Review of the BCI Competition IV},
  author={Tangermann, Michael and Müller, Klaus-Robert and Aertsen, Ad and Birbaumer, Niels and Braun, Christoph and Brunner, Clemens and Leeb, Robert and Mehring, Carsten and Miller, Kai J. and Müller-Putz, Gernot R. and Nolte, Guido and Pfurtscheller, Gert and Preissl, Hubert and Schalk, Gerwin and Schlögl, Alois and Vidaurre, Carmen and Waldert, Stephan and Blankertz, Benjamin},
  journal={Frontiers in Neuroscience},
  volume={6},
  pages={55},
  year={2012}
}

@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}
}