NTN

class NTN(*, embedding_dim: int = 100, num_slices: int = 4, non_linearity: str | Module | type[Module] | None = None, non_linearity_kwargs: Mapping[str, Any] | None = None, entity_initializer: str | Callable[[Tensor], Tensor] | None = None, **kwargs)[source]

Bases: ERModel[Tensor, tuple[Tensor, Tensor, Tensor, Tensor, Tensor], Tensor]

An implementation of NTN from [socher2013].

NTN represents entities using a \(d\)-dimensional vector. Relations are represented by

  • a \(k \times d \times d\)-dimensional tensor, \(\mathbf{W} \in \mathbb{R}^{k \times d \times d}\),

  • a \(2k \times d\)-dimensional matrix, \(\mathbf{V} \in \mathbb{R}^{k \times 2d}\), and

  • two \(k\)-dimensional vectors, \(\mathbf{b}, \mathbf{u} \in \mathbb{R}^{k}\).

Denoting the number of entities by \(E\) and the number of relations by \(R\), the total number of parameters is thus given by

\[dE + k(d^2 + 2d + 2)R\]

All representations are stored as Embedding. NTNInteraction is used as interaction upon those representations.

Note

We split the original \(k \times 2d\)-dimensional \(\mathbf{V}\) matrix into two parts of shape \(k \times d\) to support more efficient 1:n scoring, e.g., in the score_h() or score_t() setting.

See also

Initialize NTN.

Parameters:

  • embedding_dim (int) – The entity embedding dimension \(d\). Is usually \(d \in [50, 350]\).

  • num_slices (int) – The number of slices in the parameters

  • non_linearity (str | Module | type[Module] | None) – A non-linear activation function. Defaults to the hyperbolic tangent torch.nn.Tanh.

  • non_linearity_kwargs (Mapping[str, Any] | None) – If the non_linearity is passed as a class, these keyword arguments are used during its instantiation.

  • entity_initializer (str | Callable[[Tensor], Tensor] | None) – Entity initializer function. Defaults to torch.nn.init.uniform_()

  • kwargs – Remaining keyword arguments to forward to ERModel

Note

The parameter pair (non_linearity, non_linearity_kwargs) is used for class_resolver.contrib.torch.activation_resolver

An explanation of resolvers and how to use them is given in https://class-resolver.readthedocs.io/en/latest/.

Attributes Summary

hpo_default

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

Attributes Documentation

hpo_default: ClassVar[Mapping[str, Any]] = {'embedding_dim': {'high': 256, 'low': 16, 'q': 16, 'type': <class 'int'>}, 'num_slices': {'high': 4, 'low': 2, 'type': <class 'int'>}}

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