Postproduction

The postproduction pipeline takes the trigger catalogs produced by pycWB search jobs and produces final analysis products: background estimates, ranked candidate lists, detection efficiency curves, and HTML summary reports.

This section covers the complete postproduction workflow and each analysis component.

• Post-Production Workflow Guide
  • What The Workflow Consumes
  • YAML Structure
  • Variables, References, And Temporary Files
  • Recommended Complete Workflow
  • Changing Chunks Or Fractions
  • Action Reference
  • Validation And Debugging
  • Outputs To Expect
• Background Estimation
  • Overview
  • Background Estimation Method
  • Zero-Lag Identification
  • Train/FAR Data Splitting
  • FAR Lookup Table
  • Zero-Lag Significance
  • Blind Analysis (Fake Open Box)
  • Config & CLI
  • Validation Checks
• XGBoost Classification
  • Overview
  • Why XGBoost?
  • Input Features
  • Training Configuration
  • Model Output
  • Inference (Scoring)
  • Performance Considerations
  • Feature Importance
  • Validation Checks
• Training Set Preparation
  • Overview
  • Data Requirements
  • Training Set Splitting
  • Multi-Chunk Training
  • Best Practices
• Detection Efficiency
  • Overview
  • Computing Efficiency
  • Efficiency Workflow Steps
  • hrss50 and hrss90
  • Efficiency by Waveform Type
  • Efficiency vs. Sky Location
  • Visualization
  • Configurable Thresholds
  • Interpreting Efficiency Results
  • Validation Checks

Quick Start

A complete postproduction run is driven by a workflow YAML:

pycwb post-process path/to/postprocess_workflow.yaml

To inspect the dependency graph without running:

pycwb post-process path/to/postprocess_workflow.yaml --diagram-only

A reference template is available at examples/postproduction/standard_analysis_10pct_workflow.yaml.

Postproduction Architecture

The postproduction system is built on a YAML-driven workflow engine (pycwb.post_production.workflow) that chains actions as a directed acyclic graph (DAG). Actions are Python functions registered with the action_spec() decorator (pycwb.post_production.action_spec), declaring their inputs, outputs, and arguments.

Key modules:

Module	Purpose
pycwb.modules.postprocess.far	FAR vs. ranking statistic computation
pycwb.modules.postprocess.train_xgboost	XGBoost classifier training
pycwb.modules.postprocess.evaluate	Model scoring, FAR evaluation, efficiency
pycwb.modules.postprocess.selection	Trigger/job selection and train/FAR splitting
pycwb.modules.postprocess.matching	Trigger-to-injection matching
pycwb.modules.postprocess.zero_lag	Zero-lag significance analysis
pycwb.modules.postprocess.report_builder	Multi-tab HTML report generation

Typical Workflow Steps

A complete postproduction analysis follows this sequence:

1. Split background into training and FAR-holdout subsets (Training Set Preparation).

2. Match and filter simulation triggers to injection truth.

3. Train XGBoost ranking model on BKG + SIM features (XGBoost Classification).

4. Score background holdout with the trained model.

5. Build FAR lookup table from scored background (Background Estimation).

6. Score simulations and compute detection efficiency (Detection Efficiency).

7. Analyze zero-lag candidates and compute Poisson significance.

8. Generate HTML report with all results.

See Post-Production Workflow Guide for detailed YAML examples of each step.