---

name: single-table description: "Use SDV single-table synthesizers to fit, sample, conditionally sample, customize, save/load, and inspect GaussianCopula, CTGAN, TVAE, CopulaGAN, DayZ parameters, and legacy SingleTablePreset." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

SDV Single-Table Synthesis Router

Use this sub-skill when the task is to fit a single-table synthesizer, sample rows, conditionally sample rows, complete missing columns from known values, save/load a fitted model, inspect learned distributions or loss values, or choose among SDV single-table models.

Route Here For

• GaussianCopulaSynthesizer, CTGANSynthesizer, TVAESynthesizer, CopulaGANSynthesizer.
• Public DayZSynthesizer.create_parameters / validate_parameters for single-table DayZ parameter files.
• Legacy sdv.lite.SingleTablePreset, especially FAST_ML migration to GaussianCopula.
• Common synthesizer methods: fit, preprocess, fit_processed_data, sample, sample_from_conditions, sample_remaining_columns, save, load, get_info, get_parameters, get_metadata.
• Synthesizer-level constraints, transformer assignment, deep-model enable_gpu, CTGAN/TVAE loss values, GaussianCopula/CopulaGAN learned distributions.

Route Elsewhere

• Metadata detection, column sdtype editing, demo data, local CSV/Excel loading, or metadata visualization: use the data-preparation sub-skill.
• Constraint class design, programmable constraint internals, and custom constraint factories: use the constraints sub-skill; return here to attach constraints to a synthesizer.
• Quality reports and synthetic-vs-real plots: use the evaluation sub-skill.
• Relational/multi-table or sequential/time-series synthesis: use the multi-table or sequential sub-skill.

Read Order

1. For model choice, signatures, public methods, and persistence details, read references/api-reference.md.
2. For task recipes, read references/workflows.md.
3. If sampling, fitting, GPU, constraints, or DayZ parameters fail, read references/troubleshooting.md.

Operating Assumptions

• real_data is a pandas DataFrame and metadata is an SDV Metadata object that describes exactly one table. If either is missing or invalid, prepare it before selecting a model.
• Prefer the modern unified Metadata class. SingleTableMetadata and SingleTablePreset remain usable but emit deprecation warnings.
• Keep synthetic generation code self-contained: do not depend on source checkout examples or test fixtures.