Runnable walkthroughs of tsauditor. Each is self-contained; from the repository
root, install the library plus the example extras (scikit-learn for the model
comparison, jupyter/matplotlib for the notebooks):
pip install -e ".[dev,examples]"| Example | Format | Shows |
|---|---|---|
ogdc_leakage_case/ |
script + notebook | The flagship LEK001 case — a same-day feature (ChangeP) that reproduces the Direction target — on real OGDC equity data, plus the measured accuracy collapse once the leak is removed. |
beyond_price_direction/ |
notebook | Proof tsauditor is column-agnostic: VAL001/VAL002 validity on real volume, RSI, and OHLC columns — not just price/direction. |
timesfm_adapter/ |
notebook | The TimesFM adapter: audit + repair + format a raw series into a finite float32 array, with the finiteness guard and context truncation. |
sensor-example/ |
notebook + script | Structural and anomaly checks (ANO001, PRF002) on a synthetic sensor stream, and a to_pdf report demo. |
new_features_walkthrough.ipynb |
notebook | LEK004 as-of leakage, validity checks (VAL001/VAL002), tsa.fix(), and the TimesFM adapter, end-to-end. |
validation_comparison/ |
notebook | Time-series-aware validation vs. a general profiler (pandas baseline) — dependency-free. |
Several notebooks are generated by builder scripts so they stay reproducible and can be re-executed to verify they are clean:
python examples/build_new_features_notebook.py # -> new_features_walkthrough.ipynb
python examples/ogdc_leakage_case/build_ogdc_notebook.py # -> ogdc_leakage_case/ogdc_leakage.ipynb
python examples/beyond_price_direction/build_beyond_notebook.py # -> beyond_price_direction/beyond_price_direction.ipynb
python examples/timesfm_adapter/build_timesfm_notebook.py # -> timesfm_adapter/timesfm_adapter.ipynb
python examples/validation_comparison/build_validation_notebook.py # -> validation_comparison/...ipynbEach writes the .ipynb and executes every cell; a failure means the example is out of
sync with the library.