README.md

metadata

license: cc-by-4.0
task_categories:
  - time-series-forecasting
tags:
  - tsfile
  - timeseries
  - time-series
  - anomaly-detection
pretty_name: SMD (TsFile)
size_categories:
  - 100K<n<1M
configs:
  - config_name: default
    data_files:
      - split: train
        path: SMD_train.tsfile
      - split: test
        path: SMD_test.tsfile

SMD (TsFile)

Apache TsFile version of the SMD anomaly-detection subset of thuml/Time-Series-Library.

Overview

Server Machine Dataset: 38-channel machine telemetry for anomaly detection.

• Train: 708,405 rows (all normal).
• Test: 708,420 rows (with per-timestep 0/1 anomaly labels).
• Channels: 38.

The train and test segments are stored as two separate TsFiles (SMD_train.tsfile / SMD_test.tsfile), preserving the original split.

Schema (TsFile structure)

• Time (INT64, milliseconds) — row index * 1000 ms (the .npy source has no timestamp; a monotone 1 Hz proxy axis).
• FIELD (38 channels, FLOAT) — the sensor/metric channels.
• label (INT32) — per-timestep anomaly flag (0/1). The train file is all 0 (no ground-truth labels); the test file carries the anomaly labels.

No channels or rows are dropped.

Usage

Read the .tsfile files with the Apache TsFile Java or Python SDK.

Source & license

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