https://github.com/weizhepei/CasRel中run.py解读

`run.py` 是 `CasRel` 项目的入口文件，用于训练和测试模型。以下是 `run.py` 的主要代码解读和功能说明：

### 导入依赖包和模块

首先，`run.py` 导入了所需的依赖包和模块，包括 `torch`、`numpy`、`argparse`、`logging` 等。

import argparse
import logging
import os
import random
import time

import numpy as np
import torch
from torch.utils.data import DataLoader, RandomSampler, SequentialSampler

from casrel import CasRel
from dataset import RE_Dataset
from utils import init_logger, load_tokenizer, set_seed, collate_fn

### 解析命令行参数

接下来，`run.py` 解析了命令行参数，包括训练数据路径、模型保存路径、预训练模型路径、学习率等参数。

def set_args():
    parser = argparse.ArgumentParser()

    parser.add_argument("--train_data", default=None, type=str, required=True,
                        help="The input training data file (a text file).")
    parser.add_argument("--dev_data", default=None, type=str, required=True,
                        help="The input development data file (a text file).")
    parser.add_argument("--test_data", default=None, type=str, required=True,
                        help="The input testing data file (a text file).")

    parser.add_argument("--model_path", default=None, type=str, required=True,
                        help="Path to save, load model")
    parser.add_argument("--pretrain_path", default=None, type=str,
                        help="Path to pre-trained model")
    parser.add_argument("--vocab_path", default=None, type=str, required=True,
                        help="Path to vocabulary")

    parser.add_argument("--batch_size", default=32, type=int,
                        help="Batch size per GPU/CPU for training.")
    parser.add_argument("--gradient_accumulation_steps", default=1, type=int,
                        help="Number of updates steps to accumulate before performing a backward/update pass.")
    parser.add_argument("--learning_rate", default=5e-5, type=float,
                        help="The initial learning rate for Adam.")
    parser.add_argument("--num_train_epochs", default=3, type=int,
                        help="Total number of training epochs to perform.")
    parser.add_argument("--max_seq_length", default=256, type=int,
                        help="The maximum total input sequence length after tokenization. Sequences longer "
                             "than this will be truncated, sequences shorter will be padded.")
    parser.add_argument("--warmup_proportion", default=0.1, type=float,
                        help="Linear warmup over warmup_steps.")
    parser.add_argument("--weight_decay", default=0.01, type=float,
                        help="Weight decay if we apply some.")
    parser.add_argument("--adam_epsilon", default=1e-8, type=float,
                        help="Epsilon for Adam optimizer.")
    parser.add_argument("--max_grad_norm", default=1.0, type=float,
                        help="Max gradient norm.")
    parser.add_argument("--logging_steps", type=int, default=500,
                        help="Log every X updates steps.")
    parser.add_argument("--save_steps", type=int, default=500,
                        help="Save checkpoint every X updates steps.")
    parser.add_argument("--seed", type=int, default=42,
                        help="random seed for initialization")
    parser.add_argument("--device", type=str, default="cuda" if torch.cuda.is_available() else "cpu",
                        help="selected device (default: cuda if available)")

    args = parser.parse_args()

    return args

### 加载数据和模型

接下来，`run.py` 加载了训练、验证和测试数据，以及 `CasRel` 模型。

def main():
    args = set_args()

    init_logger()
    set_seed(args)

    tokenizer = load_tokenizer(args.vocab_path)

    train_dataset = RE_Dataset(args.train_data, tokenizer, args.max_seq_length)
    dev_dataset = RE_Dataset(args.dev_data, tokenizer, args.max_seq_length)
    test_dataset = RE_Dataset(args.test_data, tokenizer, args.max_seq_length)

    train_sampler = RandomSampler(train_dataset)
    train_dataloader = DataLoader(train_dataset, sampler=train_sampler, batch_size=args.batch_size,
                                  collate_fn=collate_fn)

    dev_sampler = SequentialSampler(dev_dataset)
    dev_dataloader = DataLoader(dev_dataset, sampler=dev_sampler, batch_size=args.batch_size,
                                collate_fn=collate_fn)

    test_sampler = SequentialSampler(test_dataset)
    test_dataloader = DataLoader(test_dataset, sampler=test_sampler, batch_size=args.batch_size,
                                  collate_fn=collate_fn)

    model = CasRel(args)
    if args.pretrain_path:
        model.load_state_dict(torch.load(args.pretrain_path, map_location="cpu"))
        logging.info(f"load pre-trained model from {args.pretrain_path}")

    model.to(args.device)

### 训练模型

接下来，`run.py` 开始训练模型，包括前向传播、反向传播、梯度更新等步骤。

    optimizer = torch.optim.Adam([{'params': model.bert.parameters(), 'lr': args.learning_rate},
                                  {'params': model.subject_fc.parameters(), 'lr': args.learning_rate},
                                  {'params': model.object_fc.parameters(), 'lr': args.learning_rate},
                                  {'params': model.predicate_fc.parameters(), 'lr': args.learning_rate},
                                  {'params': model.linear.parameters(), 'lr': args.learning_rate}],
                                 lr=args.learning_rate, eps=args.adam_epsilon, weight_decay=args.weight_decay)

    total_steps = len(train_dataloader) // args.gradient_accumulation_steps * args.num_train_epochs
    warmup_steps = int(total_steps * args.warmup_proportion)

    scheduler = torch.optim.lr_scheduler.LambdaLR(
        optimizer,
        lr_lambda=lambda epoch: 1 / (1 + 0.05 * (epoch - 1))
    )

    global_step = 0
    best_f1 = 0
    for epoch in range(args.num_train_epochs):
        for step, batch in enumerate(train_dataloader):
            model.train()
            batch = tuple(t.to(args.device) for t in batch)

            inputs = {
                "input_ids": batch[0],
                "attention_mask": batch[1],
                "token_type_ids": batch[2],
                "subj_pos": batch[3],
                "obj_pos": batch[4],
                "subj_type": batch[5],
                "obj_type": batch[6],
                "subj_label": batch[7],
                "obj_label": batch[8],
                "predicate_label": batch[9],
            }

            outputs = model(**inputs)
            loss = outputs[0]
            loss.backward()

            if (step + 1) % args.gradient_accumulation_steps == 0:
                torch.nn.utils.clip_grad_norm_(model.parameters(), args.max_grad_norm)

                optimizer.step()
                scheduler.step()
                optimizer.zero_grad()
                global_step += 1

                if global_step % args.logging_steps == 0:
                    logging.info(f"Epoch:[{epoch + 1}]/[{args.num_train_epochs}] Step:[{global_step}] "
                                 f"Train loss:{loss.item():.6f}")

                if global_step % args.save_steps == 0:
                    f1 = evaluate(model, dev_dataloader, args)
                    if f1 > best_f1:
                        best_f1 = f1
                        torch.save(model.state_dict(), os.path.join(args.model_path, "best_model.bin"))
                        logging.info(f"Save model at step [{global_step}] with best f1 {best_f1:.4f}")

### 测试模型

最后，`run.py` 对模型进行测试，输出模型在测试集上的预测结果。

    model.load_state_dict(torch.load(os.path.join(args.model_path, "best_model.bin"), map_location="cpu"))
    logging.info(f"load best model from {os.path.join(args.model_path, 'best_model.bin')}")

    f1, precision, recall = evaluate(model, test_dataloader, args)

    logging.info(f"Test f1:{f1:.4f} precision:{precision:.4f} recall:{recall:.4f}")

以上就是 `run.py` 的主要代码解读和功能说明。