A conditional-GAN-style architecture designed for tabular data with mixed continuous/categorical columns and multi-modal distributions. It learns to generate rows that a discriminator cannot reliably separate from real ones. Best default for larger, more complex datasets.
Matflow Hybrid
Combines a GAN-style network with a copula transform of the margins. The copula anchors the dependence structure, which makes training more stable on small samples than a pure neural model.
Matflow Statistical
Fits parametric marginal distributions plus a correlation structure directly — no neural network involved. Fastest mode, most defensible when rows number in the tens rather than hundreds, and a useful control against the neural modes.
Rule of thumb: < 100 rows → Statistical first. 100–500 → Hybrid or Statistical. > 500 → Neural. Then always confirm with Evaluation.