teras.losses.ctgan_generator_loss

teras.losses.ctgan_generator_loss#

teras.losses.ctgan_generator_loss(x_generated, y_pred_generated, cond_vectors, mask, metadata)[source]#

Loss for the Generator model in the CTGAN architecture.

CTGAN is a state-of-the-art tabular data generation architecture proposed by Lei Xu et al. in the paper, “Modeling Tabular data using Conditional GAN”.

Reference(s):

https://arxiv.org/abs/1907.00503

Parameters:

  • x_generated – Samples drawn from the input dataset

  • y_pred_generated – Discriminator’s output for the generated samples

  • cond_vectors – Conditional vectors that are used for and with generated samples

  • mask – Mask created during the conditional vectors generation step

  • metadata – dict, metadata computed during the data transformation step.

Returns:

Generator’s loss.

Contents