teras.layers module

Offers layers used in building the state of the art tabular deep learning architectures available in teras.

CategoricalEmbedding(embedding_dim, ...)	Categorical Embedding layer that create trainable embeddings for categorical features values.
CategoricalExtraction(categorical_idx, **kwargs)	CategoricalExtraction layer extracts categorical features from inputs as is.
CLSToken(embedding_dim, **kwargs)	CLS Token layer that makes it possible to append CLS token embedding to the input embeddings in the sequential or functional models.
CLSTokenExtraction([axis])	Extracts CLS Token embeddings.
ContinuousExtraction(continuous_idx, **kwargs)	ContinuousExtraction layer extracts continuous features from inputs as is.
CutMix([probability, mask_seed, shuffle_seed])	CutMix is a regularization layer proposed by Sangdoo Yun et al. in the paper, CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features.
CTGANGeneratorLayer([dim])	Residual Block for Generator as used by the authors of CTGAN proposed in the paper Modeling Tabular data using Conditional GAN.
CTGANDiscriminatorLayer([dim, ...])	Discriminator Layer based on the architecture proposed by Lei Xu et al. in the paper, "Modeling Tabular data using Conditional GAN".
FTTransformerFeatureTokenizer(cardinalities, ...)	Feature Tokenizer layer based on FT-Transformer architecture proposed in the "Revisiting Deep Learning Models for Tabular Data" paper.
GumbelSoftmax([temperature, hard, seed])	Implementation of the Gumbel Softmax activation proposed by Eric Jang et al. in the paper, "Categorical Reparameterization with Gumbel-Softmax".
LayerList(layers[, sequential])	LayerList is a list of layers, but is also a layer itself.
MixUp([alpha, lambda_seed, shuffle_seed])	MixUp is a regularization layer proposed by Hongyi Zhang et al. in the paper, mixup: BEYOND EMPIRICAL RISK MINIMIZATION.
SAINTEmbedding(embedding_dim, cardinalities, ...)	SAINTEmbedding layer as proposed in the paper, "SAINT: Improved Neural Networks for Tabular Data".
SAINTEncoderLayer(embedding_dim[, ...])	SAINTEncoderLayer layer as proposed in the paper, "SAINT: Improved Neural Networks for Tabular Data".
SAINTMultiHeadInterSampleAttention(...[, ...])	Multi Head Inter Sample Attention layer based on the SAINT architecture proposed in the "SAINT: Improved Neural Networks for Tabular Data" paper.
SAINTProjectionHead(hidden_dim, output_dim)	Projection Head layer that is used in the contrastive learning phase of the SAINTPretrainer to project embeddings to a lower dimension.
SAINTReconstructionHead(cardinalities, ...)	SAINT Reconstruction Head layer for SAINTPretrainer.
TabTransformerColumnEmbedding(cardinalities, ...)	Column Embedding layer as proposed in the "TabTransformer: Tabular Data Modeling Using Contextual Embeddings".
TabNetFeatureTransformerLayer(dim, ...)	TabNetFeatureTransformerLayer layer that serves as the building block for the TabNetFeatureTransformer layer which is proposed by Arik et al. in the "TabNet: Attentive Interpretable Tabular Learning" paper.
TabNetAttentiveTransformer(data_dim, ...)	TabNetAttentiveTransformer layer proposed by Arik et al. in the "TabNet: Attentive Interpretable Tabular Learning" paper.
TransformerEncoderLayer(embedding_dim[, ...])	Transformer Encoder Layer as proposed in the original Transformer architecture in the "Attention is all you need" paper.
TransformerFeedForward(embedding_dim[, ...])	Transformer Feed Forward layer as proposed in the original Transformers architecture, in the paper,"Attention is all you need", with a slight addition of optional Dropout layer.