DistMult-WN18RR-single-gpu-adv-wandb¶

这一部分介绍如何用一个 GPU 在 WN18RR 知识图谱上训练 DistMult [YYH+15]，应用 RotatE [SDNT19] 提出的自我对抗负采样损失函数进行模型训练，使用 wandb 记录实验结果。

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.utils import WandbLogger
from pybind11_ke.config import Trainer, Tester
from pybind11_ke.module.model import DistMult
from pybind11_ke.module.loss import SigmoidLoss
from pybind11_ke.module.strategy import NegativeSampling
from pybind11_ke.data import TrainDataLoader, TestDataLoader

首先初始化 pybind11_ke.utils.WandbLogger 日志记录器，它是对 wandb 初始化操作的一层简单封装。

wandb_logger = WandbLogger(
    project="pybind11-ke",
    name="distmult",
    config=dict(
            in_path = "../../benchmarks/WN18RR/",
            batch_size = 2000,
            threads = 1,
            sampling_mode = "cross",
            bern = False,
            neg_ent = 64,
            neg_rel = 0,
            dim = 1024,
            adv_temperature = 0.5,
            l3_regul_rate = 0.000005,
            use_gpu = True,
            device = 'cuda:1',
            epochs = 400,
            lr = 0.002,
            opt_method = "adam",
            test = True,
            valid_interval = 100,
            log_interval = 100,
            save_interval = 100,
            save_path = '../../checkpoint/distmult.pth'
    )
)

config = wandb_logger.config

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 = config.in_path,
    batch_size = config.batch_size,
    threads = config.threads,
    sampling_mode = config.sampling_mode,
    bern = config.bern,
    neg_ent = config.neg_ent,
    neg_rel = config.neg_rel
)

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.DistMult，它是最简单的双线性模型。

# define the model
distmult = DistMult(
    ent_tol = train_dataloader.get_ent_tol(),
    rel_tol = train_dataloader.get_rel_tol(),
    dim = config.dim
)

损失函数¶

我们这里使用了逻辑损失函数：pybind11_ke.module.loss.SigmoidLoss， pybind11_ke.module.strategy.NegativeSampling 对 pybind11_ke.module.loss.SigmoidLoss 进行了封装，加入权重衰减等额外项。除此之外，我们使用 adv_temperature 开启了 RotatE 提出的自我对抗负采样。

# define the loss function
model = NegativeSampling(
    model = distmult,
    loss = SigmoidLoss(adv_temperature = config.adv_temperature),
    batch_size = train_dataloader.get_batch_size(),
    l3_regul_rate = config.l3_regul_rate
)

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。

# dataloader for test
test_dataloader = TestDataLoader(config.in_path)

# test the model
tester = Tester(model = distmult, data_loader = test_dataloader, use_gpu = config.use_gpu, device = config.device)

# train the model
trainer = Trainer(model = model, data_loader = 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, use_wandb = True)
trainer.run()

# close your wandb run
wandb_logger.finish()