Skip to contents

Returns a p x n matrix of realistic mediator feature vectors drawn from the feature-level copula texture model. Each row is a feature (transcript), each column a sample. The draws preserve the marginal distributions and cross-feature correlation structure of the user's mediator panel. The texture is drawn independently of U and never fed into X, so it cannot open a confounding backdoor.

Usage

.iteration_mediator_texture_features(trained_gan, n, p)

Arguments

trained_gan

An iconic_gan object (may be NULL).

n

Number of synthetic samples.

p

Number of features (target n_features for this iteration).

Value

A p x n matrix, or NULL.

Details

When no feature_texture is available (trained_gan is NULL or lacks $feature_texture), returns NULL and the caller falls back to scalar noise.