forked from uma-pi1/kge
-
Notifications
You must be signed in to change notification settings - Fork 0
/
rescal.py
95 lines (79 loc) · 3 KB
/
rescal.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
import torch
from kge import Config, Dataset
from kge.model.kge_model import KgeEmbedder, KgeModel, RelationalScorer
class RescalScorer(RelationalScorer):
r"""Implementation of the RESCAL KGE scorer."""
def __init__(self, config: Config, dataset: Dataset, configuration_key=None):
super().__init__(config, dataset, configuration_key)
def score_emb(
self,
s_emb: torch.Tensor,
p_emb: torch.Tensor,
o_emb: torch.Tensor,
combine: str,
):
batch_size = p_emb.size(0)
entity_size = s_emb.size(-1)
# reshape relation embeddings to obtain mixing matrices for RESCAL
p_mixmat = p_emb.view(-1, entity_size, entity_size)
if combine == "spo":
out = (
s_emb.unsqueeze(1) # [batch x 1 x entity_size]
.bmm(p_mixmat) # apply mixing matrices
.view(batch_size, entity_size) # drop dim 1
* o_emb # apply object embeddings
).sum(
dim=-1
) # and sum to obtain predictions
elif combine == "sp_":
out = (
s_emb.unsqueeze(1)
.bmm(p_mixmat)
.view(batch_size, entity_size)
.mm(o_emb.transpose(0, 1))
)
elif combine == "_po":
out = (
p_mixmat.bmm(o_emb.unsqueeze(2))
.view(batch_size, entity_size)
.mm(s_emb.transpose(0, 1))
)
else:
return super().score_emb(s_emb, p_emb, o_emb, combine)
return out.view(batch_size, -1)
class Rescal(KgeModel):
r"""Implementation of the RÉSCAL KGE model."""
def __init__(
self,
config: Config,
dataset: Dataset,
configuration_key=None,
init_for_load_only=False,
):
self._init_configuration(config, configuration_key)
rescal_set_relation_embedder_dim(
config, dataset, self.configuration_key + ".relation_embedder"
)
super().__init__(
config=config,
dataset=dataset,
scorer=RescalScorer,
configuration_key=self.configuration_key,
init_for_load_only=init_for_load_only,
)
def rescal_set_relation_embedder_dim(config, dataset, rel_emb_conf_key):
"""Set the relation embedder dimensionality for RESCAL in the config.
If <0, set it to the square of the size of the entity embedder. Else leave
unchanged.
"""
dim = config.get_default(rel_emb_conf_key + ".dim")
if dim < 0: # autodetect relation embedding dimensionality
ent_emb_conf_key = rel_emb_conf_key.replace(
"relation_embedder", "entity_embedder"
)
if ent_emb_conf_key == rel_emb_conf_key:
raise ValueError(
"Cannot determine relation embedding size; please set manually."
)
dim = config.get_default(ent_emb_conf_key + ".dim") ** 2
config.set(rel_emb_conf_key + ".dim", dim, log=True)