DAGMMDetector<T>

Models & Types · AiDotNet.AnomalyDetection.NeuralNetwork

Implements DAGMM (Deep Autoencoding Gaussian Mixture Model) for anomaly detection.

Inheritance: ModelBase<T, Matrix<T>, Vector<T>> → AnomalyDetectorBase<T> → DAGMMDetector<T>

Implements: IGeneratedParameterRegistrar<T>

For Beginners

DAGMM combines an autoencoder with a Gaussian Mixture Model in an end-to-end trainable architecture. The autoencoder learns compressed representations, and the GMM learns the density of normal data in the latent space.

How It Works

The algorithm works by:

  1. Compression network (autoencoder) learns latent representation
  2. Estimation network predicts GMM membership probabilities
  3. GMM models the distribution of latent codes + reconstruction features
  4. Anomaly score is the negative log-likelihood under the GMM

When to use:

  • Complex multivariate data
  • When you want to model multiple modes of normal behavior
  • When reconstruction error alone is insufficient

Industry Standard Defaults:

  • Latent dimensions: 4
  • Hidden dimensions: 64
  • Number of mixtures: 4
  • Epochs: 100
  • Contamination: 0.1 (10%)

Reference: Zong, B., et al. (2018). “Deep Autoencoding Gaussian Mixture Model for Unsupervised Anomaly Detection.” ICLR.

Constructors

Constructor Summary
DAGMMDetector(int latentDim, int hiddenDim, int numMixtures, int epochs, double learningRate, double contamination, int randomSeed) Creates a new DAGMM anomaly detector.

Properties

Property Summary
int LatentDim { get; } Gets the latent dimensions.
int NumMixtures { get; } Gets the number of GMM components.

Methods

Method Summary
Vector<T> ScoreAnomalies(Matrix<T> X)
void Fit(Matrix<T> X)