备注
Go to the end to download the full example code
ANALOGY-WN18RR-single-gpu || ANALOGY-WN18RR-single-gpu-wandb || ANALOGY-WN18RR-single-gpu-hpo
ANALOGY-WN18RR-single-gpu¶
这一部分介绍如何用一个 GPU 在 WN18RR 知识图谱上训练 ANALOGY [LWY17]。
导入数据¶
pybind11-OpenKE 有两个工具用于导入数据: pybind11_ke.data.TrainDataLoader 和
pybind11_ke.data.TestDataLoader。
from pybind11_ke.config import Trainer, Tester
from pybind11_ke.module.model import Analogy
from pybind11_ke.module.loss import SoftplusLoss
from pybind11_ke.module.strategy import NegativeSampling
from pybind11_ke.data import TrainDataLoader, TestDataLoader
pybind11-KE 提供了很多数据集,它们很多都是 KGE 原论文发表时附带的数据集。
pybind11_ke.data.TrainDataLoader 包含 in_path 用于传递数据集目录。
# dataloader for training
train_dataloader = TrainDataLoader(
in_path = "../../benchmarks/WN18RR/",
nbatches = 100,
threads = 1,
sampling_mode = "normal",
bern = True,
neg_ent = 25,
neg_rel = 0
)
导入模型¶
pybind11-OpenKE 提供了很多 KGE 模型,它们都是目前最常用的基线模型。我们下面将要导入
pybind11_ke.module.model.Analogy,它是双线性模型的集大成者。
# define the model
analogy = Analogy(
ent_tol = train_dataloader.get_ent_tol(),
rel_tol = train_dataloader.get_rel_tol(),
dim = 200
)
损失函数¶
我们这里使用了逻辑损失函数:pybind11_ke.module.loss.SoftplusLoss,
pybind11_ke.module.strategy.NegativeSampling 对
pybind11_ke.module.loss.SoftplusLoss 进行了封装,加入权重衰减等额外项。
# define the loss function
model = NegativeSampling(
model = analogy,
loss = SoftplusLoss(),
batch_size = train_dataloader.get_batch_size(),
regul_rate = 1.0
)
训练模型¶
pybind11-OpenKE 将训练循环包装成了 pybind11_ke.config.Trainer,
可以运行它的 pybind11_ke.config.Trainer.run() 函数进行模型学习;
也可以通过传入 pybind11_ke.config.Tester,
使得训练器能够在训练过程中评估模型;pybind11_ke.config.Tester 使用
pybind11_ke.data.TestDataLoader 作为数据采样器。
# dataloader for test
test_dataloader = TestDataLoader('../../benchmarks/WN18RR/')
# test the model
tester = Tester(model = analogy, data_loader = test_dataloader, use_gpu = True, device = 'cuda:1')
# train the model
trainer = Trainer(model = model, data_loader = train_dataloader,
epochs = 2000, lr = 0.5, opt_method = "adagrad", use_gpu = True, device = 'cuda:1',
tester = tester, test = True, valid_interval = 100,
log_interval = 100, save_interval = 100,
save_path = '../../checkpoint/analogy.pth', delta = 0.01)
trainer.run()
Total running time of the script: ( 0 minutes 0.000 seconds)