Datasets:

THUgewu
/

Handwriting

license: cc-by-4.0
task_categories:
  - time-series-forecasting
tags:
  - tsfile
  - timeseries
  - time-series
  - classification
pretty_name: Handwriting (TsFile)
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: Handwriting_train.tsfile
      - split: test
        path: Handwriting_test.tsfile
      - split: train_labels
        path: labels/Handwriting_train_labels.csv
      - split: test_labels
        path: labels/Handwriting_test_labels.csv

Handwriting (TsFile)

Apache TsFile version of the Handwriting UEA classification subset of thuml/Time-Series-Library.

Overview

Accelerometer traces of writing the 26 letters; classify the letter.

Train samples: 150.
Test samples: 850.
Dimensions (channels): 3.
Series length: 152.
Classes: 26.

Each sample is an independent multivariate series. TRAIN and TEST are stored as two separate TsFiles (Handwriting_train.tsfile / Handwriting_test.tsfile). Class labels are stored as separate CSV sidecar files in labels/.

Schema (TsFile structure)

sample_index (TAG) - one device per sample. Query one sample with WHERE sample_index=0.
Time (INT64) - within-sample position (0..151), a frame index rather than a wall-clock timestamp.
dim_0..dim_2 (FIELD, FLOAT) - the 3 channels at each position.
Labels (CSV sidecar) - labels/Handwriting_train_labels.csv and labels/Handwriting_test_labels.csv contain sample_index,class_label, one row per sample.

Converted from the sktime .ts format. No dimensions, time points, samples, or labels are dropped. Labels are not stored inside the TsFile feature tables.

Usage

Read the .tsfile files with the Apache TsFile Java or Python SDK. Join labels from the sidecar CSV files by sample_index within the matching split.

Source & license

Original dataset: https://huggingface.co/datasets/thuml/Time-Series-Library (subset Handwriting)
Author / publisher: thuml (Tsinghua University)
Paper: https://arxiv.org/abs/2407.13278
License: CC BY 4.0