TabNetClassifier<T>

Models & Types · AiDotNet.NeuralNetworks.Tabular

TabNet implementation for classification tasks.

Inheritance: TabNetBase<T> → TabNetClassifier<T>

For Beginners

Use TabNetClassifier when you want to categorize data into different classes (groups) from tabular data.

Example use cases:

  • Customer churn prediction (will stay / will leave)
  • Fraud detection (legitimate / fraudulent transaction)
  • Medical diagnosis (healthy / disease A / disease B / …)
  • Credit risk assessment (low / medium / high risk)

Key features:

  • Automatic Feature Selection: Learns which features matter for classification
  • Interpretability: See which features drove each classification decision
  • Multi-class Support: Works for binary (2 classes) or multi-class problems
  • Probability Outputs: Returns confidence scores for each class

Basic usage:

How It Works

TabNetClassifier extends the TabNet architecture for predicting discrete class labels. It uses softmax activation on the output layer to produce class probabilities.

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

Constructors

Constructor Summary
TabNetClassifier(int inputFeatures, int numClasses, TabNetOptions<T> options) Initializes a new instance of the TabNetClassifier class.

Methods

Method Summary
T ComputeAccuracy(int[] predictions, Tensor<T> targets) Computes the accuracy (percentage of correct predictions).
T ComputeCrossEntropyLoss(Tensor<T> probabilities, Tensor<T> targets) Computes the cross-entropy loss for classification.
T ComputeCrossEntropyLossOneHot(Tensor<T> probabilities, Tensor<T> oneHotTargets) Computes cross-entropy loss with one-hot encoded targets.
T ComputeF1Score(int[] predictions, Tensor<T> targets, int positiveClass) Computes the F1 score for binary classification.
T ComputeTotalLoss(Tensor<T> probabilities, Tensor<T> targets) Computes the total loss including sparsity regularization.
T TrainStep(Tensor<T> input, Tensor<T> targets, T learningRate) Performs a single training step.
Tensor<T> ComputeCrossEntropyGradient(Tensor<T> probabilities, Tensor<T> targets) Computes the gradient of cross-entropy loss with softmax.
Tensor<T> ComputeCrossEntropyGradientOneHot(Tensor<T> probabilities, Tensor<T> oneHotTargets) Computes the gradient with one-hot encoded targets.
Tensor<T> Forward(Tensor<T> input) Performs forward pass with softmax activation.
Tensor<T> ForwardLogits(Tensor<T> input) Gets the raw logits without softmax.
Tensor<T> PredictProbabilities(Tensor<T> input) Gets class probabilities for each sample.
int[] Predict(Tensor<T> input) Predicts class labels (indices of highest probability classes).