pRotatE

pRotatE scoring function.

Attributes

• name

Examples

>>> from ckb import models
>>> from ckb import datasets
>>> from ckb import scoring

>>> import torch

>>> _ = torch.manual_seed(42)

>>> dataset = datasets.Semanlink(1)

>>> model = models.DistillBert(
...    entities = dataset.entities,
...    relations = dataset.relations,
...    gamma = 9,
...    device = 'cpu',
...    scoring = scoring.pRotatE(),
... )

>>> sample = torch.tensor([[0, 0, 0], [2, 2, 2]])
>>> model(sample)
tensor([[5.4199],
        [5.4798]], grad_fn=<ViewBackward>)

>>> sample = torch.tensor([[0, 0, 1], [2, 2, 1]])
>>> model(sample)
tensor([[5.4521],
        [5.5248]], grad_fn=<ViewBackward>)

>>> sample = torch.tensor([[1, 0, 0], [1, 2, 2]])
>>> model(sample)
tensor([[5.4962],
        [5.5059]], grad_fn=<ViewBackward>)

>>> sample = torch.tensor([[0, 0, 0], [2, 2, 2]])
>>> negative_sample = torch.tensor([[1, 0], [1, 2]])

>>> model(sample, negative_sample, mode='head-batch')
tensor([[5.4962, 5.4199],
        [5.5059, 5.4798]], grad_fn=<ViewBackward>)

>>> model(sample, negative_sample, mode='tail-batch')
tensor([[5.4521, 5.4199],
        [5.5248, 5.4798]], grad_fn=<ViewBackward>)

Methods

call

Compute the score of given facts (heads, relations, tails).

Parameters

• head
• relation
• tail
• gamma
• embedding_range
• modulus
• mode
• kwargs