TabTransformerGenGenerator<T>

Models & Types · AiDotNet.NeuralNetworks.SyntheticData

TabTransformer-Gen generator that uses column-wise contextual embeddings and masked prediction to generate realistic tabular data.

Inheritance: NeuralNetworkBase<T> → NeuralSyntheticTabularGeneratorBase<T> → TabTransformerGenGenerator<T>

Implements: ISyntheticTabularGenerator<T>

For Beginners

Think of this like a crossword puzzle:

  • Each column is a clue that helps fill in other columns
  • During training, we hide some columns and learn to fill them in
  • During generation, we start with an empty puzzle and fill in one column at a time,

using already-filled columns as context for the remaining ones

If you provide custom layers in the architecture, those will be used for the feed-forward network blocks. If not, the network creates standard TabTransformer layers based on the original paper specifications.

Example usage:

How It Works

TabTransformer-Gen treats each column as a “token” in a sequence:

Training: Mask random columns, predict their values from the unmasked context. Generation: Start with all columns masked, iteratively unmask (predict) columns.

This implementation follows the standard neural network architecture pattern with:

  • Proper inheritance from NeuralNetworkBase
  • Layer-based architecture using ILayer components
  • Full forward/backward/update lifecycle

Reference: “TabTransformer: Tabular Data Modeling Using Contextual Embeddings” (Huang et al., 2020) — adapted for generation with masked prediction

Constructors

Constructor Summary
TabTransformerGenGenerator() Initializes a new TabTransformer-Gen generator with the specified architecture.
TabTransformerGenGenerator(NeuralNetworkArchitecture<T> architecture, TabTransformerGenOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction)

Properties

Property Summary
IReadOnlyList<ColumnMetadata> Columns { get; }
TabTransformerGenOptions<T> TransformerOptions { get; } Gets the TabTransformerGen-specific options.
bool IsFitted { get; set; }

Methods

Method Summary
Dictionary<string, T> GetFeatureImportance()
Matrix<T> Generate(int numSamples, Vector<T> conditionColumn, Vector<T> conditionValue)
ModelMetadata<T> GetModelMetadata()
Task FitAsync(Matrix<T> data, IReadOnlyList<ColumnMetadata> columns, int? epochs, CancellationToken ct)
Tensor<T> ForwardForTraining(Tensor<T> input) The training forward — identical to Predict — overridden so the tape-based training path runs the column-token transformer rather than the default sequential walk over Layers (which would feed each layer’s output to the next, ignoring t…
void Fit(Matrix<T> data, IReadOnlyList<ColumnMetadata> columns, int? epochs)