OCTGANOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for OCT-GAN (Kim et al. 2021, “OCT-GAN: Neural ODE-based Conditional Tabular GANs”), a conditional tabular GAN whose generator maps z ⊕ c through a neural ODE and whose discriminator classifies the trajectory of a neural ODE at learnable time points.

Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → FinancialNeuralNetworkOptions → RiskModelOptions<T> → SyntheticDataGeneratorOptions<T> → OCTGANOptions<T>

For Beginners

OCT-GAN learns to produce new table rows that look like the real ones. You can ask it for rows with a given category in a discrete column (the “condition”), and it learns to honour that request.

Example:

How It Works

Defaults follow section 4.1.4 of the paper: 256-wide hidden layers in both networks, m = 3 trajectory segments, one shared learning rate λ = λ_G = λ_D = λ_t = 2e-4 from the paper’s grid (the time points share the critic’s optimizer, as in the authors’ code), a decay factor ψ = 0.97 every ξ = 2 epochs, 300 epochs, a mini-batch of 500 and 10 VGM modes.

Constructors

Constructor Summary
OCTGANOptions()

Properties

Property Summary
double DiscriminatorDropout { get; set; } Gets or sets the dropout rate of the discriminator’s first two layers (Eq.
double GradientPenaltyWeight { get; set; } Gets or sets the gradient-penalty weight λ of the WGAN-GP loss (Eq.
double GumbelTemperature { get; set; } Gets or sets the Gumbel-softmax temperature of the generator’s mode and discrete-column heads (Eqs.
double LearningRate { get; set; } Gets or sets the generator and discriminator learning rate λ_G = λ_D.
double LearningRateDecay { get; set; } Gets or sets the decay factor ψ applied to λ_G and λ_D (and so to the time points) every LearningRateDecayPeriod epochs.
int BatchSize { get; set; } Gets or sets the training batch size.
int DiscriminatorSteps { get; set; } Gets or sets the number of discriminator updates per generator update.
int EmbeddingDimension { get; set; } Gets or sets the dimension of the noise vector z.
int Epochs { get; set; } Gets or sets the number of training epochs.
int LearningRateDecayPeriod { get; set; } Gets or sets the decay period ξ in epochs.
int OdeSolverSteps { get; set; } Gets or sets the number of fourth-order Runge-Kutta steps used for each ODE integral (the generator’s [0, 1] and each of the discriminator’s [t_{i-1}, t_i]).
int TimePoints { get; set; } Gets or sets the number m of segments the discriminator’s ODE trajectory on [0, 1] is read at: t_0 = 0 and t_m = 1 are fixed and the m − 1 interior points t_1..t_{m-1} are learned.
int VGMModes { get; set; } Gets or sets the number of VGM modes for continuous column encoding.
int[] DiscriminatorDimensions { get; set; } Gets or sets the output widths of the discriminator’s layers before its ODE (FC1 and FC2, Eq.
int[] GeneratorDimensions { get; set; } Gets or sets the output widths of the generator’s two residual blocks (FC1 and FC2, Eqs.