# coding:utf-8
#
# unike/module/model/HolE.py
#
# git pull from OpenKE-PyTorch by LuYF-Lemon-love <luyanfeng_nlp@qq.com> on May 7, 2023
# updated by LuYF-Lemon-love <luyanfeng_nlp@qq.com> on Jan 31, 2024
#
# 该头文件定义了 HolE.
"""
HolE - 利用循环相关进行知识图谱嵌入，是 RESCAL 的压缩版本，因此非常容易的应用于大型的知识图谱。
"""
import torch
import typing
import torch.nn as nn
from .Model import Model
from typing_extensions import override
[文档]class HolE(Model):
"""
``HolE`` :cite:`HolE` 提出于 2016 年，利用循环相关进行知识图谱嵌入，是 RESCAL 的压缩版本，因此非常容易的应用于大型的知识图谱。
评分函数为:
.. math::
\mathbf{r}^T (\mathcal{F}^{-1}(\overline{\mathcal{F}(\mathbf{h})} \odot \mathcal{F}(\mathbf{t})))
其中 :math:`\mathcal{F}(\cdot)` 和 :math:`\mathcal{F}^{-1}(\cdot)` 表示快速傅里叶变换，:math:`\overline{\mathbf{x}}` 表示复数共轭，:math:`\odot` 表示哈达玛积。
正三元组的评分函数的值越大越好，负三元组越小越好，如果想获得更详细的信息请访问 :ref:`HolE <hole>`。
例子::
from unike.utils import WandbLogger
from unike.data import KGEDataLoader, BernSampler, TradTestSampler
from unike.module.model import HolE
from unike.module.loss import SoftplusLoss
from unike.module.strategy import NegativeSampling
from unike.config import Trainer, Tester
wandb_logger = WandbLogger(
project="unike",
name="HolE-WN18RR",
config=dict(
in_path = '../../benchmarks/WN18RR/',
batch_size = 8192,
neg_ent = 25,
test = True,
test_batch_size = 256,
num_workers = 16,
dim = 100,
regul_rate = 1.0,
use_gpu = True,
device = 'cuda:0',
epochs = 1000,
lr = 0.5,
opt_method = 'adagrad',
valid_interval = 100,
log_interval = 100,
save_interval = 100,
save_path = '../../checkpoint/hole.pth'
)
)
config = wandb_logger.config
# dataloader for training
dataloader = KGEDataLoader(
in_path = config.in_path,
batch_size = config.batch_size,
neg_ent = config.neg_ent,
test = config.test,
test_batch_size = config.test_batch_size,
num_workers = config.num_workers,
train_sampler = BernSampler,
test_sampler = TradTestSampler
)
# define the model
hole = HolE(
ent_tol = dataloader.get_ent_tol(),
rel_tol = dataloader.get_rel_tol(),
dim = config.dim
)
# define the loss function
model = NegativeSampling(
model = hole,
loss = SoftplusLoss(),
regul_rate = config.regul_rate
)
# test the model
tester = Tester(model = hole, data_loader = dataloader, use_gpu = config.use_gpu, device = config.device)
# train the model
trainer = Trainer(model = model, data_loader = dataloader.train_dataloader(), epochs = config.epochs,
lr = config.lr, opt_method = config.opt_method, use_gpu = config.use_gpu, device = config.device,
tester = tester, test = config.test, valid_interval = config.valid_interval,
log_interval = config.log_interval, save_interval = config.save_interval,
save_path = config.save_path, wandb_logger = wandb_logger)
trainer.run()
# close your wandb run
wandb_logger.finish()
"""
[文档] def __init__(
self,
ent_tol: int,
rel_tol: int,
dim: int = 100):
"""创建 HolE 对象。
:param ent_tol: 实体的个数
:type ent_tol: int
:param rel_tol: 关系的个数
:type rel_tol: int
:param dim: 实体和关系嵌入向量的维度
:type dim: int
"""
super(HolE, self).__init__(ent_tol, rel_tol)
#: 实体和关系嵌入向量的维度
self.dim: int = dim
#: 根据实体个数，创建的实体嵌入
self.ent_embeddings: torch.nn.Embedding = nn.Embedding(self.ent_tol, self.dim)
#: 根据关系个数，创建的关系嵌入
self.rel_embeddings: torch.nn.Embedding = nn.Embedding(self.rel_tol, self.dim)
nn.init.xavier_uniform_(self.ent_embeddings.weight.data)
nn.init.xavier_uniform_(self.rel_embeddings.weight.data)
[文档] @override
def forward(
self,
triples: torch.Tensor,
negs: torch.Tensor = None,
mode: str = 'single') -> torch.Tensor:
"""
定义每次调用时执行的计算。
:py:class:`torch.nn.Module` 子类必须重写 :py:meth:`torch.nn.Module.forward`。
:param triples: 正确的三元组
:type triples: torch.Tensor
:param negs: 负三元组类别
:type negs: torch.Tensor
:param mode: 模式
:type triples: str
:returns: 三元组的得分
:rtype: torch.Tensor
"""
head_emb, relation_emb, tail_emb = self.tri2emb(triples, negs, mode)
score = self._calc(head_emb, relation_emb, tail_emb)
return score
[文档] def _calc(
self,
h: torch.Tensor,
r: torch.Tensor,
t: torch.Tensor) -> torch.Tensor:
"""计算 HolE 的评分函数。
:param h: 头实体的向量。
:type h: torch.Tensor
:param r: 关系的向量。
:type r: torch.Tensor
:param t: 尾实体的向量。
:type t: torch.Tensor
:returns: 三元组的得分
:rtype: torch.Tensor
"""
score = self._ccorr(h, t) * r
score = torch.sum(score, -1)
return score
[文档] def _ccorr(
self,
a: torch.Tensor,
b: torch.Tensor) -> torch.Tensor:
"""计算循环相关 :math:`\mathcal{F}^{-1}(\overline{\mathcal{F}(\mathbf{h})} \odot \mathcal{F}(\mathbf{t}))`。
利用 :py:func:`torch.fft.rfft` 计算实数到复数离散傅里叶变换，:py:func:`torch.fft.irfft` 是其逆变换；
利用 :py:func:`torch.conj` 计算复数的共轭。
:param a: 头实体的向量。
:type a: torch.Tensor
:param b: 尾实体的向量。
:type b: torch.Tensor
:returns: 返回循环相关计算结果。
:rtype: torch.Tensor
"""
# 计算傅里叶变换
a_fft = torch.fft.rfft(a, dim=-1)
b_fft = torch.fft.rfft(b, dim=-1)
# 复数的共轭
a_fft = torch.conj(a_fft)
# 哈达玛积
p_fft = a_fft * b_fft
# 傅里叶变换的逆变换
return torch.fft.irfft(p_fft, n=a.shape[-1], dim=-1)
[文档] @override
def predict(
self,
data: dict[str, typing.Union[torch.Tensor,str]],
mode) -> torch.Tensor:
"""HolE 的推理方法。
:param data: 数据。
:type data: dict[str, typing.Union[torch.Tensor,str]]
:returns: 三元组的得分
:rtype: torch.Tensor
"""
triples = data["positive_sample"]
head_emb, relation_emb, tail_emb = self.tri2emb(triples, mode=mode)
score = self._calc(head_emb, relation_emb, tail_emb)
return score
[文档] def regularization(
self,
data: dict[str, typing.Union[torch.Tensor, str]]) -> torch.Tensor:
"""L2 正则化函数（又称权重衰减），在损失函数中用到。
:param data: 数据。
:type data: dict[str, typing.Union[torch.Tensor, str]]
:returns: 模型参数的正则损失
:rtype: torch.Tensor
"""
pos_sample = data["positive_sample"]
neg_sample = data["negative_sample"]
mode = data["mode"]
pos_head_emb, pos_relation_emb, pos_tail_emb = self.tri2emb(pos_sample)
if mode == "bern":
neg_head_emb, neg_relation_emb, neg_tail_emb = self.tri2emb(neg_sample)
else:
neg_head_emb, neg_relation_emb, neg_tail_emb = self.tri2emb(pos_sample, neg_sample, mode)
pos_regul = (torch.mean(pos_head_emb ** 2) +
torch.mean(pos_relation_emb ** 2) +
torch.mean(pos_tail_emb ** 2)) / 3
neg_regul = (torch.mean(neg_head_emb ** 2) +
torch.mean(neg_relation_emb ** 2) +
torch.mean(neg_tail_emb ** 2)) / 3
regul = (pos_regul + neg_regul) / 2
return regul
[文档] def l3_regularization(self) -> torch.Tensor:
"""L3 正则化函数，在损失函数中用到。
:returns: 模型参数的正则损失
:rtype: torch.Tensor
"""
return (self.ent_embeddings.weight.norm(p = 3)**3 + self.rel_embeddings.weight.norm(p = 3)**3)
[文档]def get_hole_hpo_config() -> dict[str, dict[str, typing.Any]]:
"""返回 :py:class:`HolE` 的默认超参数优化配置。
默认配置为::
parameters_dict = {
'model': {
'value': 'HolE'
},
'dim': {
'values': [50, 100, 200]
}
}
:returns: :py:class:`HolE` 的默认超参数优化配置
:rtype: dict[str, dict[str, typing.Any]]
"""
parameters_dict = {
'model': {
'value': 'HolE'
},
'dim': {
'values': [50, 100, 200]
}
}
return parameters_dict