QuatE

class QuatE(*, embedding_dim: int = 100, entity_initializer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function init_quaternions>, entity_regularizer: str | ~pykeen.regularizers.Regularizer | None = <class 'pykeen.regularizers.LpRegularizer'>, entity_regularizer_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, relation_initializer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function init_quaternions>, relation_regularizer: str | ~pykeen.regularizers.Regularizer | None = <class 'pykeen.regularizers.LpRegularizer'>, relation_regularizer_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, relation_normalizer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function normalize>, **kwargs)[source]

Bases: ERModel[Tensor, Tensor, Tensor]

An implementation of QuatE from [zhang2019].

QuatE uses hypercomplex valued representations for the entities and relations. Entities and relations are represented as vectors \(\textbf{e}_i, \textbf{r}_i \in \mathbb{H}^d\), and the plausibility score is computed using the quaternion inner product.

The representations are stored in an Embedding. Scores are calculated with QuatEInteraction.

See also

Official implementation: https://github.com/cheungdaven/QuatE/blob/master/models/QuatE.py

Initialize QuatE.

Note

The default parameters correspond to the first setting for FB15k-237 described from [zhang2019].

Parameters:

  • embedding_dim (int) –

    The embedding dimensionality of the entity embeddings.

    Note

    The number of parameter per entity is 4 * embedding_dim, since quaternion are used.

  • entity_initializer (str | Callable[[Tensor], Tensor] | None) – The initializer to use for the entity embeddings.

  • entity_regularizer (str | Regularizer | None) – The regularizer to use for the entity embeddings.

  • entity_regularizer_kwargs (Mapping[str, Any] | None) – The keyword arguments passed to the entity regularizer. Defaults to QuatE.regularizer_default_kwargs if not specified.

  • relation_initializer (str | Callable[[Tensor], Tensor] | None) – The initializer to use for the relation embeddings.

  • relation_regularizer (str | Regularizer | None) – The regularizer to use for the relation embeddings.

  • relation_regularizer_kwargs (Mapping[str, Any] | None) – The keyword arguments passed to the relation regularizer. Defaults to QuatE.regularizer_default_kwargs if not specified.

  • relation_normalizer (str | Callable[[Tensor], Tensor] | None) – The normalizer to use for the relation embeddings.

  • kwargs – Additional keyword based arguments passed to pykeen.models.ERModel. Must not contain “interaction”, “entity_representations”, or “relation_representations”.

Attributes Summary

hpo_default

The default strategy for optimizing the model's hyper-parameters

loss_default_kwargs

The default parameters for the default loss function class

regularizer_default_kwargs

The LP settings used by [zhang2019] for QuatE.

Attributes Documentation

hpo_default: ClassVar[Mapping[str, Any]] = {'embedding_dim': {'high': 256, 'low': 16, 'q': 16, 'type': <class 'int'>}}

The default strategy for optimizing the model’s hyper-parameters

loss_default_kwargs: ClassVar[Mapping[str, Any]] = {'reduction': 'mean'}

The default parameters for the default loss function class

regularizer_default_kwargs: ClassVar[Mapping[str, Any]] = {'normalize': True, 'p': 2.0, 'weight': 0.029999999999999992}

The LP settings used by [zhang2019] for QuatE.