TripleREInteraction

class TripleREInteraction(u: float | None = 1.0, p: int = 1, power_norm: bool = False)[source]

Bases: NormBasedInteraction[Tensor, tuple[Tensor, Tensor, Tensor], Tensor]

The TripleRE interaction function from [yu2021].

It is given by

\[\mathbf{h} \odot (\mathbf{r}_h + u) - \mathbf{t} \odot (\mathbf{r}_t + u) + \mathbf{r}\]

with head entity, relation and tail entity representations \(\mathbf{h}, \mathbf{r}, \mathbf{t} \in \mathbb{R}^d\), relation specific head and tail multipliers \(\mathbf{r}_h, \mathbf{r}_t \in \mathbb{R}^d\), and a scalar relation factor offset \(u \in \mathbb{R}\).

Note

This interaction is equivalent to LineaREInteraction except the \(u\) term. The \(u\) is only non-zero for the version 2 from the paper.

Note

For equivalence to the paper version, h and t should be normalized to unit Euclidean length, and p and power_norm be kept at their default values.

Initialize the module.

See also

The parameter p and power_norm are directly passed to NormBasedInteraction.

Parameters:
  • u (float | None) – The relation factor offset. Can be set to None (or 0) to disable it.

  • p (int) – The norm used with torch.linalg.vector_norm(). Typically is 1 or 2.

  • power_norm (bool) – Whether to use the p-th power of the \(L_p\) norm. It has the advantage of being differentiable around 0, and numerically more stable.

Attributes Summary

relation_shape

The symbolic shapes for relation representations

Methods Summary

forward(h, r, t)

Evaluate the interaction function.

Attributes Documentation

relation_shape: Sequence[str] = ('d', 'd', 'd')

The symbolic shapes for relation representations

Methods Documentation

forward(h: Tensor, r: tuple[Tensor, Tensor, Tensor], t: Tensor) Tensor[source]

Evaluate the interaction function.

See also

Interaction.forward for a detailed description about the generic batched form of the interaction function.

Parameters:
  • h (Tensor) – shape: (*batch_dims, d) The head representations.

  • r (tuple[Tensor, Tensor, Tensor]) – shape: (*batch_dims, d), 3 times The relation representations.

  • t (Tensor) – shape: (*batch_dims, d) The tail representations.

Returns:

shape: batch_dims The scores.

Return type:

Tensor