teras.losses.tabnet_reconstruction_loss

teras.losses.tabnet_reconstruction_loss(real=None, reconstructed=None, mask=None)

Reconstruction loss for TabNet Pretrainer mode as proposed by Sercan et al. in the paper, “TabNet: Attentive Interpretable Tabular Learning”

Reference(s):

https://arxiv.org/abs/1908.07442

Parameters:

• real – Samples drawn from the input dataset

• reconstructed – Samples reconstructed by the decoder

• mask – Mask that indicates the missing-ness of features in a sample

Returns:

Reconstruction loss for TabNet Pretraining.