TabNetBase<T>

Base Classes · AiDotNet.NeuralNetworks.Tabular

Base implementation of the TabNet architecture for attentive interpretable tabular learning.

Known derived types: TabNetClassifier<T>, TabNetRegression<T>

For Beginners

TabNet is a neural network designed specifically for tables of data (like spreadsheets or databases). Unlike traditional neural networks that process all features at once, TabNet makes decisions in steps:

  1. Step 1: “Which features should I look at first?”
  • Selects a subset of features using sparse attention
  • Processes those features through the Feature Transformer
  1. Step 2: “Based on what I learned, which features next?”
  • Uses information from Step 1 to select new features
  • Features used before are less likely to be selected again
  1. Steps 3, 4, …: Continue refining the decision

At each step, TabNet accumulates knowledge that gets combined into the final output.

This approach provides several benefits:

  • Interpretability: You can see which features were selected at each step
  • Feature Selection: Automatically ignores irrelevant features
  • Efficiency: Doesn’t need to process all features for every prediction
  • Performance: Often matches or beats gradient boosting on tabular data

How It Works

TabNet is a deep learning architecture specifically designed for tabular data. It uses sequential attention to choose which features to reason from at each decision step, enabling both high performance and interpretability.

Reference: “TabNet: Attentive Interpretable Tabular Learning” (Arik & Pfister, AAAI 2021)

Properties

Property Summary
long ParameterCount { get; } Gets the total number of trainable parameters.

Methods

Method Summary
IReadOnlyList<Tensor<T>> GetStepAttentionMasks() Gets the attention masks from each decision step.
T ComputeSparsityLoss() Computes the sparsity regularization loss.
Tensor<T> Forward(Tensor<T> input) Performs the forward pass through the TabNet encoder.
Tensor<T> GetFeatureImportance() Gets the feature importance from all decision steps.
Vector<T> GetParameters() Gets all trainable parameters.
void ResetState() Resets the internal state.
void SetParameters(Vector<T> parameters) Sets the trainable parameters.
void SetTrainingMode(bool isTraining) Sets training mode.
void UpdateParameters(T learningRate) Updates parameters using the specified learning rate.