BayesianNetworkSynthOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for Bayesian Network Synthesis, a statistical approach that learns a directed acyclic graph (DAG) structure and conditional probability tables to generate synthetic tabular data via ancestral sampling.

Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → FinancialNeuralNetworkOptions → RiskModelOptions<T> → BayesianNetworkSynthOptions<T>

For Beginners

This method creates a probabilistic model of your data:

Think of a family tree of features — some features “depend on” others. For example, in a health dataset:

  1. Age has no parents (sampled first)
  2. Blood pressure depends on Age
  3. Medication depends on Blood pressure

The model learns these dependency chains and samples new data following the same parent-to-child order, producing statistically coherent rows.

Unlike neural network generators (CTGAN, TVAE), this uses classical statistics, making it faster to train and more interpretable, though less flexible for complex distributions.

Example:

How It Works

With differential privacy enabled (the default) this is PrivBayes (Zhang et al. 2017):

  • Structure learning: GreedyBayes picks each attribute’s parents with the exponential mechanism
  • Parameter estimation: d - k joint distributions get Laplace noise; conditionals derive from them
  • Ancestral sampling: Generates data by sampling from root nodes down through the DAG

With it disabled, the structure is learned by greedy BIC hill-climbing and the conditional tables are Laplace-smoothed maximum-likelihood estimates (no privacy guarantee).

Constructors

Constructor Summary
BayesianNetworkSynthOptions() Initializes a new instance with the default settings.
BayesianNetworkSynthOptions(BayesianNetworkSynthOptions<T> other) Initializes a new instance by copying every setting of another instance.

Properties

Property Summary
Vector<T> DomainLowerBounds { get; set; } Gets or sets the public lower bound of each column’s domain, used for the equal-width bins.
Vector<T> DomainUpperBounds { get; set; } Gets or sets the public upper bound of each column’s domain, used for the equal-width bins.
bool EnableDifferentialPrivacy { get; set; } Gets or sets whether to enforce differential privacy, which is what makes this PrivBayes rather than a plain Bayesian-network synthesizer.
double LaplaceSmoothing { get; set; } Gets or sets the Laplace smoothing constant for CPT estimation.
double PrivacyBudget { get; set; } Gets or sets the total privacy budget, epsilon.
double StructureBudgetFraction { get; set; } Gets or sets the fraction of the total privacy budget spent on learning the network structure, with the remainder spent on noising the conditional distributions.
double UsefulnessThreshold { get; set; } Gets or sets PrivBayes’ usefulness threshold theta, used to choose the network degree k.
int MaxIterations { get; set; } Gets or sets the maximum number of structure learning iterations.
int MaxParents { get; set; } Gets or sets the maximum number of parents per node in the DAG (the upper bound on PrivBayes’ degree k).
int NumBins { get; set; } Gets or sets the number of discretization bins for continuous features.