TabSynOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for TabSyn, a state-of-the-art synthetic tabular data generator that combines a VAE with latent diffusion for high-quality generation.

Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → FinancialNeuralNetworkOptions → RiskModelOptions<T> → SyntheticDataGeneratorOptions<T> → TabSynOptions<T>

For Beginners

TabSyn is like TVAE + TabDDPM combined in a clever way:

  1. First, a VAE learns to “compress” your data into a small summary (latent codes)
  2. Then, a diffusion model learns the distribution of those compressed summaries
  3. To generate new data: the diffusion model creates new summaries, and the VAE’s

decoder converts them back to realistic data rows

This two-step approach often produces the highest quality synthetic data because:

  • The VAE handles the complex mixed-type structure
  • The diffusion model learns a simpler, continuous distribution in latent space

Example:

How It Works

TabSyn operates in two phases:

  1. VAE pretraining: Learns a compact latent representation of the tabular data
  2. Latent diffusion: Trains a diffusion model in the VAE’s latent space

Generation: Sample from the diffusion model in latent space, then decode with the VAE.

Reference: “TabSyn: Bridging the Gap” (Zhang et al., NeurIPS 2023)

Constructors

Constructor Summary
TabSynOptions()

Properties

Property Summary
double BetaEnd { get; set; } Gets or sets the beta schedule end value for diffusion.
double BetaStart { get; set; } Gets or sets the beta schedule start value for diffusion.
double DiffusionLearningRate { get; set; } Gets or sets the learning rate for the diffusion model.
double VAELearningRate { get; set; } Gets or sets the learning rate for the VAE.
int BatchSize { get; set; } Gets or sets the training batch size.
int DiffusionEpochs { get; set; } Gets or sets the number of epochs for diffusion model training.
int DiffusionSteps { get; set; } Gets or sets the number of diffusion timesteps.
int LatentDimension { get; set; } Gets or sets the dimension of the VAE latent space.
int TimestepEmbeddingDimension { get; set; } Gets or sets the dimension of the timestep embedding for the diffusion model.
int VAEEpochs { get; set; } Gets or sets the number of epochs for VAE pretraining.
int VGMModes { get; set; } Gets or sets the number of VGM modes for continuous column normalization.
int[] DecoderDimensions { get; set; } Gets or sets the hidden layer sizes for the VAE decoder.
int[] DiffusionMLPDimensions { get; set; } Gets or sets the hidden layer sizes for the diffusion denoiser MLP.
int[] EncoderDimensions { get; set; } Gets or sets the hidden layer sizes for the VAE encoder.