Reporter YAML

The Reporter module is responsible for outputting experiment results and supports multiple report formats including data storage, evaluation reports, and timing records.

Usage Examples

Click the button below to run examples in Colab:

Note: If using Colab, please see the runtime setup guide.

Loader:
  load_data:
    filepath: benchmark://adult-income
    schema: benchmark://adult-income_schema
Synthesizer:
  synthesize:
    method: default
Evaluator:
  fidelity_check:
    method: sdmetrics-qualityreport

Reporter:
  save_synthetic:
    method: save_data
    source: Synthesizer

  save_report:
    method: save_report
    granularity: global

  save_schema:
    method: save_schema
    source:
      - Loader
      - Synthesizer

  save_timing:
    method: save_timing

Supported Report Methods

This module supports the following report output methods. For detailed parameters, please refer to each subpage:

1. Save Data - Save synthetic data or outputs from other modules as CSV files
2. Generate Evaluation Reports - Generate evaluation result reports with multiple granularity levels
3. Save Schema Information - Output data column schema information from specified modules
4. Save Validation Results - Export Constrainer validation results as structured CSV reports
5. Save Timing Information - Record execution time of each module

Naming Strategy and Experiment Names

PETsARD adopts a unified experiment naming convention for identifying and tracking experiment processes.

Naming Strategy Overview

The Reporter module supports two naming strategies, controlled via the naming_strategy parameter:

1. TRADITIONAL: Maintains backward compatibility with traditional naming format
2. COMPACT: Provides a more concise and readable naming format

Experiment Name Format

Experiment Tuple

full_expt_tuple is a tuple composed of module name and experiment name:

(module_name, experiment_name)

This format is primarily used by the Reporter system to identify and organize experiment results.

Experiment String

full_expt_name concatenates the module name and experiment name with a hyphen:

{module_name}-{experiment_name}

File Naming Format

save_data Method

save_report Method

save_timing Method

All strategies use the same format:

{output}_timing_report.csv

Module Abbreviation Table (Compact Strategy)

Granularity Abbreviation Table (Compact Strategy)

Naming Recommendations

1. Module Names

   • Use standard module names: ‘Synthesizer’, ‘Evaluator’, ‘Processor’, etc.
   • Case sensitivity must match exactly

2. Experiment Names

   • Use meaningful prefixes: ’exp’, ’eval’, ’test’, etc.
   • Separate different parts with underscores: method name, parameter settings, etc.
   • Use square brackets for evaluation levels: [global], [columnwise], [pairwise]

3. Parameter Encoding

   • Use abbreviations for parameter names: method, batch, epoch, etc.
   • Use concise representations for values: 300, 0.1, etc.
   • Connect multiple parameters with underscores: method_a_batch500

4. Strategy Selection

   • New Projects: Recommended to use compact strategy for more concise filenames
   • Existing Projects: Use traditional strategy to maintain compatibility
   • File Management: compact strategy makes filenames easier to sort and categorize

Common Parameters

All report methods support the following common parameters:

• output (string, optional) - Output file name prefix, defaults to petsard
• naming_strategy (string, optional) - Filename naming strategy, choose traditional (default) or compact

Notes

• Reporter should be executed after all modules requiring reports have completed
• Files with the same name will be overwritten; use different output prefixes
• All reports are saved in CSV format with UTF-8 encoding
• Recommended to use compact naming strategy for new projects; use traditional for existing projects to ensure compatibility