RotatE

class RotatE(*, embedding_dim: int = 200, entity_initializer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function xavier_uniform_>, relation_initializer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function init_phases>, relation_constrainer: str | ~typing.Callable[[~torch.Tensor], ~torch.Tensor] | None = <function complex_normalize>, regularizer: str | ~pykeen.regularizers.Regularizer | type[~pykeen.regularizers.Regularizer] | None = None, regularizer_kwargs: ~collections.abc.Mapping[str, ~typing.Any] | None = None, **kwargs)[source]

Bases: ERModel[Tensor, Tensor, Tensor]

An implementation of RotatE from [sun2019].

RotatE models relations as rotations from head to tail entities in complex space:

\[\textbf{e}_t= \textbf{e}_h \odot \textbf{r}_r\]

where \(\textbf{e}, \textbf{r} \in \mathbb{C}^{d}\) and the complex elements of \(\textbf{r}_r\) are restricted to have a modulus of one (\(\|\textbf{r}_r\| = 1\)). The interaction model is then defined as:

\[f(h,r,t) = -\|\textbf{e}_h \odot \textbf{r}_r - \textbf{e}_t\|\]

which allows to model symmetry, antisymmetry, inversion, and composition.