Implementation:Sdv dev SDV CTGANSynthesizer Init

Knowledge Sources	SDV CTGAN
Domains	Deep_Learning, Synthetic_Data
Last Updated	2026-02-14 00:00 GMT

Overview

Concrete tool for creating a CTGAN-based synthesizer for single-table synthetic data generation, provided by the SDV library.

Description

The CTGANSynthesizer wraps the ctgan.CTGAN model. It provides configurable neural network architecture (embedding, generator, and discriminator dimensions), training hyperparameters (learning rates, epochs, batch size), and GPU support.

Usage

Import this class when you need to generate single-table synthetic data using a deep learning approach, particularly for datasets with complex inter-column relationships.

Code Reference

Source Location

Repository: SDV
File: sdv/single_table/ctgan.py
Lines: L109-229

Signature

class CTGANSynthesizer(LossValuesMixin, MissingModuleMixin, BaseSingleTableSynthesizer):
    def __init__(
        self,
        metadata,
        enforce_min_max_values=True,
        enforce_rounding=True,
        locales=['en_US'],
        embedding_dim=128,
        generator_dim=(256, 256),
        discriminator_dim=(256, 256),
        generator_lr=2e-4,
        generator_decay=1e-6,
        discriminator_lr=2e-4,
        discriminator_decay=1e-6,
        batch_size=500,
        discriminator_steps=1,
        log_frequency=True,
        verbose=False,
        epochs=300,
        pac=10,
        enable_gpu=True,
        cuda=None,
    ):
        """
        Args:
            metadata (Metadata): Single table metadata.
            enforce_min_max_values (bool): Clip to min/max. Defaults to True.
            enforce_rounding (bool): Round as original. Defaults to True.
            locales (list): Locale(s) for AnonymizedFaker. Defaults to ['en_US'].
            embedding_dim (int): Random sample size for Generator. Defaults to 128.
            generator_dim (tuple): Residual layer sizes. Defaults to (256, 256).
            discriminator_dim (tuple): Discriminator layer sizes. Defaults to (256, 256).
            generator_lr (float): Generator learning rate. Defaults to 2e-4.
            generator_decay (float): Generator weight decay. Defaults to 1e-6.
            discriminator_lr (float): Discriminator learning rate. Defaults to 2e-4.
            discriminator_decay (float): Discriminator weight decay. Defaults to 1e-6.
            batch_size (int): Batch size. Defaults to 500.
            discriminator_steps (int): Discriminator updates per generator update. Defaults to 1.
            log_frequency (bool): Use log frequency for conditional sampling. Defaults to True.
            verbose (bool): Print progress. Defaults to False.
            epochs (int): Training epochs. Defaults to 300.
            pac (int): PacGAN group size. Defaults to 10.
            enable_gpu (bool): Use GPU if available. Defaults to True.
            cuda (bool or str): Deprecated. Use enable_gpu instead.
        """

Import

from sdv.single_table import CTGANSynthesizer

I/O Contract

Inputs

Name	Type	Required	Description
metadata	Metadata	Yes	Single table metadata object
epochs	int	No	Training epochs (default: 300)
batch_size	int	No	Training batch size (default: 500)
embedding_dim	int	No	Generator input dimension (default: 128)
generator_dim	tuple	No	Generator layer sizes (default: (256, 256))
discriminator_dim	tuple	No	Discriminator layer sizes (default: (256, 256))
enable_gpu	bool	No	Attempt GPU computation (default: True)

Outputs

Name	Type	Description
instance	CTGANSynthesizer	Unfitted synthesizer ready for .fit() call

Usage Examples

Basic Usage

from sdv.datasets.demo import download_demo
from sdv.single_table import CTGANSynthesizer

data, metadata = download_demo(modality='single_table', dataset_name='fake_hotel_guests')

synthesizer = CTGANSynthesizer(metadata, epochs=100, verbose=True)
synthesizer.fit(data)
synthetic_data = synthesizer.sample(num_rows=500)

Related Pages

Implements Principle

Principle:Sdv_dev_SDV_CTGAN_Synthesis

Requires Environment

Uses Heuristic

Heuristic:Sdv_dev_SDV_CTGAN_Column_Performance

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment