Diff¶

datasetlint diff compares two datasets and classifies changes, regressions, and improvements.

datasetlint diff old_dataset new_dataset
datasetlint diff old_dataset new_dataset --format json
datasetlint diff old_dataset new_dataset --fail-on-regression

Python:

from datasetlint.diff import compare_datasets

report = compare_datasets("old_dataset", "new_dataset")
print(report.summary)

The diff covers:

• Metadata changes
• Sensors added or removed
• Duration changes
• Frame count changes
• Frequency changes
• Calibration changes
• Label class distribution changes
• Issue count changes
• Trajectory speed and acceleration summary changes

Regressions include removed sensors, significant frame-count or duration drops, calibration changes, disappeared label classes, and increased warning/error issue counts. Use --fail-on-regression in CI to exit non-zero when regressions are found.

Diff reports are printed to stdout. Redirect JSON or Markdown output when you want an artifact:

datasetlint diff old_dataset new_dataset --format json > datasetlint-diff.json

Relevant thresholds:

frame_count_drop_ratio_warning: 0.1
duration_drop_ratio_warning: 0.1
issue_regression_severity: warning