dicee.models.literal

Classes

LiteralEmbeddings

A model for learning and predicting numerical literals using pre-trained KGE.

Module Contents

class dicee.models.literal.LiteralEmbeddings(num_of_data_properties: int, embedding_dims: int, entity_embeddings: torch.tensor, dropout: float = 0.3, gate_residual=True, freeze_entity_embeddings=True)[source]

Bases: torch.nn.Module

A model for learning and predicting numerical literals using pre-trained KGE.

num_of_data_properties

Number of data properties (attributes).

Type:: int

embedding_dims

Dimension of the embeddings.

Type:: int

entity_embeddings

Pre-trained entity embeddings.

Type:: torch.tensor

dropout

Dropout rate for regularization.

Type:: float

gate_residual

Whether to use gated residual connections.

Type:: bool

freeze_entity_embeddings

Whether to freeze the entity embeddings during training.

Type:: bool

embedding_dim

num_of_data_properties

hidden_dim

gate_residual = True

freeze_entity_embeddings = True

entity_embeddings

data_property_embeddings

fc

fc_out

dropout

gated_residual_proj

layer_norm

forward(entity_idx, attr_idx)[source]

Parameters:

entity_idx (Tensor) – Entity indices (batch).
attr_idx (Tensor) – Attribute (Data property) indices (batch).

Returns:

scalar predictions.

Return type:

Tensor

property device