Skip to content

Test Offline Pipeline (Engineering View)

Role

The test offline pipelines run trained models on held-out Parquet/CSV data and produce Excel reports, confusion matrices, and diagnostic plots (e.g. size, location, leak series). They answer: “How would this model perform on new cases?”

Why It Matters for Engineering

  • By-case evaluation: Typically we evaluate per case or per file, so metrics align with the by-case split philosophy (see Train/validation split by case).
  • Threshold and metrics: Detection threshold and metric definitions (precision, recall, F1, etc.) are consistent with training and model validation.
  • World-class analysis: Reports and plots support the “world-class” analysis goal (see World-class detection analysis).

Configuration and Scripts

  • Scripts: run_test_offline_pipeline.py, run_pfm_test_offline_pipeline.py.
  • Config: parquet_root, output_report_folder, model paths, feature/window config, sampling options.
  • Detailed reference: See Test offline in the main Pipelines section.

This Engineering page gives the rationale; the full configuration and output description are in the Pipelines docs.