Module: tfr.keras.layers

Defines Keras Layers for TF-Ranking.


class Bilinear: A Keras Layer makes bilinear interaction of two vectors.

class ConcatFeatures: Concatenates context features and example features in a listwise manner.

class DocumentInteractionAttention: Cross Document Interaction Attention layer.

class FlattenList: Layer to flatten the example list.

class GAMLayer: Defines a generalized additive model (GAM) layer.

class RestoreList: Output layer to restore listwise output shape.

class SelfAttentionMask: Create 3D attention mask from a 2D tensor mask.


create_tower(...): Creates a feed-forward network as tf.keras.Sequential.