Inference

Entity Inference

class torchkge.inference.EntityInference(model, known_entities, known_relations, top_k=1, missing='tails', dictionary=None)[source]

Use trained embedding model to infer missing entities in triples.

Parameters

model (torchkge.models.interfaces.Model) – Embedding model inheriting from the right interface.
known_entities (torch.Tensor, shape: (n_facts), dtype: torch.long) – List of the indices of known entities.
known_relations (torch.Tensor, shape: (n_facts), dtype: torch.long) – List of the indices of known relations.
top_k (int) – Indicates the number of top predictions to return.
missing (str) – String indicating if the missing entities are the heads or the tails.
dictionary (dict, optional (default=None)) – Dictionary of possible heads or tails (depending on the value of missing). It is used to filter predictions that are known to be True in the training set in order to return only new facts.

model

Embedding model inheriting from the right interface.

Type: torchkge.models.interfaces.Model

known_entities

List of the indices of known entities.

Type: torch.Tensor, shape: (n_facts), dtype: torch.long

known_relations

List of the indices of known relations.

Type: torch.Tensor, shape: (n_facts), dtype: torch.long

top_k

Indicates the number of top predictions to return.

Type: int

missing

String indicating if the missing entities are the heads or the tails.

Type: str

dictionary

Dictionary of possible heads or tails (depending on the value of missing). It is used to filter predictions that are known to be True in the training set in order to return only new facts.

Type: dict, optional (default=None)

predictions

List of the indices of predicted entities for each test fact.

Type: torch.Tensor, shape: (n_facts, self.top_k), dtype: torch.long

scores

List of the scores of resulting triples for each test fact.

Type: torch.Tensor, shape: (n_facts, self.top_k), dtype: torch.float

Relation Inference

class torchkge.inference.RelationInference(model, entities1, entities2, top_k=1, dictionary=None)[source]

Use trained embedding model to infer missing relations in triples.

Parameters

model (torchkge.models.interfaces.Model) – Embedding model inheriting from the right interface.
entities1 (torch.Tensor, shape: (n_facts), dtype: torch.long) – List of the indices of known entities 1.
entities2 (torch.Tensor, shape: (n_facts), dtype: torch.long) – List of the indices of known entities 2.
top_k (int) – Indicates the number of top predictions to return.
dictionary (dict, optional (default=None)) – Dictionary of possible relations. It is used to filter predictions that are known to be True in the training set in order to return only new facts.

model

Embedding model inheriting from the right interface.

Type: torchkge.models.interfaces.Model

entities1

List of the indices of known entities 1.

Type: torch.Tensor, shape: (n_facts), dtype: torch.long

entities2

List of the indices of known entities 2.

Type: torch.Tensor, shape: (n_facts), dtype: torch.long