Skip to content

YAML Configuration Reference¶

Complete reference for all YAML configuration options.

`data` Section¶

Key	Type	Required	Description
`real`	string	Yes	Path to real CSV file
`synthetic`	string	No	Path to synthetic CSV (or `"None"`)
`target`	string	No	Target column name (or `"None"`)
`task_type`	string	No	`classification`, `regression`, or `"None"`
`real_separator`	string	No	CSV separator for real file (default: `,`)
`synth_separator`	string	No	CSV separator for synthetic file (default: `,`)
`schema`	dict	Yes	Column type definitions

`metrics` Section¶

List of metric IDs or shortcuts:

metrics:
  - "fidelity.ks"            # Single metric
  - "fidelity.marginal"      # Shortcut (all marginal)
  - "privacy"                # All privacy metrics

`reproducibility` Section¶

Key	Type	Default	Description
`seed`	int	42	Base random seed

`calibration` Section¶

Key	Type	Default	Description
`n_iterations`	int	10	Split-half calibration iterations
`sample_percentage`	float	100.0	% of dataset to use
`n_jobs`	int	1	Parallel jobs (-1 = all cores)
`tune_aggregators`	bool	true	Optimize aggregation weights

`evaluation` Section¶

Key	Type	Default	Description
`n_runs`	int	1	Number of repetitions (multi-seed)

`benchmark` Section¶

Key	Type	Default	Description
`enabled`	bool	false	Enable benchmark mode
`output_dir`	string	—	Output directory for results
`n_runs`	int	5	Seeds per generator
`sample_ratio`	float	1.0	Fraction of data to use
`generators`	list	—	Generator definitions
`statistical_test.method`	string	`friedman-nemenyi`	Test method
`statistical_test.alpha`	float	0.05	Significance level

Generator definition¶

generators:
  - name: "ctgan"            # Registry key
    params:                   # Passed as kwargs to constructor
      epochs: 300
      batch_size: 500

`report` Section¶

Key	Type	Default	Description
`formats`	list	`["json"]`	Output formats: `json`, `md`
`output_dir`	string	`reports/`	Output directory