TabDDPMOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for TabDDPM (Tabular Denoising Diffusion Probabilistic Model), a diffusion-based model for generating realistic synthetic tabular data.

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

For Beginners

TabDDPM works by gradually destroying data with noise, then learning to reverse the process. Think of it like a restoration expert:

Training (learning to restore):

  1. Take a real data row
  2. Add a random amount of noise (more noise = more destroyed)
  3. Tell the model “this was step t out of 1000” and ask it to predict the noise
  4. The model learns to undo any amount of noise

Generation (creating new data):

  1. Start with pure random noise
  2. Ask the model to remove a tiny bit of noise (step 999 to 998)
  3. Repeat 1000 times until you have a clean, realistic row

For numbers: regular Gaussian noise (like static on a TV) For categories: noise means randomly changing the category toward “equally likely”

Example:

How It Works

TabDDPM applies denoising diffusion models to tabular data with separate processes for numerical and categorical features:

  • Gaussian diffusion for continuous/numerical columns (adds Gaussian noise)
  • Multinomial diffusion for categorical columns (transitions toward uniform distribution)
  • A shared MLP denoiser with timestep embedding predicts the original data from noisy input

Reference: “TabDDPM: Modelling Tabular Data with Diffusion Models” (Kotelnikov et al., ICML 2023)

Constructors

Constructor Summary
TabDDPMOptions()

Properties

Property Summary
double BetaEnd { get; set; } Gets or sets the ending value of the beta schedule.
double BetaStart { get; set; } Gets or sets the starting value of the beta schedule.
double DropoutRate { get; set; } Gets or sets the dropout rate for the denoiser MLP.
double LearningRate { get; set; } Gets or sets the learning rate for the optimizer.
int BatchSize { get; set; } Gets or sets the training batch size.
int Epochs { get; set; } Gets or sets the number of training epochs.
int NumCategoricalDiffusionSteps { get; set; } Gets or sets the number of diffusion steps for the multinomial (categorical) diffusion process.
int NumTimesteps { get; set; } Gets or sets the number of diffusion timesteps.
int TimestepEmbeddingDimension { get; set; } Gets or sets the dimension of the timestep embedding.
int[] MLPDimensions { get; set; } Gets or sets the hidden layer sizes for the denoiser MLP.
string BetaSchedule { get; set; } Gets or sets the beta (noise variance) schedule type.