SyntheticDataGeneratorOptions<T>
Options & Configuration · AiDotNet.Models.Options
Options shared by the synthetic tabular data generators — the GAN, VAE and diffusion models in
src/NeuralNetworks/SyntheticData that learn a data distribution and sample new rows from
it.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → FinancialNeuralNetworkOptions → RiskModelOptions<T> → SyntheticDataGeneratorOptions<T>
Known derived types: AutoDiffTabOptions<T>, CTABGANPlusOptions<T>, CTGANOptions<T>, CausalGANOptions<T>, CopulaGANOptions<T>, DPCTGANOptions<T>, FinDiffOptions<T>, GOGGLEOptions<T>, MedGANOptions<T>, MisGANOptions<T>, OCTGANOptions<T>, PATEGANOptions<T>, REaLTabFormerOptions<T>, TVAEOptions<T>, TabDDPMOptions<T>, TabFlowOptions<T>, TabLLMGenOptions<T>, TabSynOptions<T>, TabTransformerGenOptions<T>, TableGANOptions<T>, TimeGANOptions<T>
For Beginners
These models generate believable fake records that look like your real data — useful when the real data cannot be shared. They train two or more networks against each other, which is a less stable process than ordinary supervised training, and that instability is what this class exists to account for.
How It Works
Why this sits between the generators and RiskModelOptions. All 21 of
these generators declared a gradient-clipping threshold of 5.0, while the twelve tabular
prediction models that share RiskModelOptions as a base declared 1.0.
They were siblings with different values, so neither number could live on the shared parent:
putting 5.0 there would have loosened clipping on the twelve, and leaving the inherited 1.0
would have tightened it on these 21. The split is exactly the
SyntheticData / Tabular folder boundary, so the family is a real one rather than
an artefact of the migration.
Constructors
| Constructor | Summary |
|---|---|
SyntheticDataGeneratorOptions() |
Initializes a new instance carrying the defaults shared by the synthetic data generators. |