Datawhale AI 夏令营 NLP 方向 Task 01 学习笔记

Task1 - 速通 Baseline

导入package

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
from torchtext.data.utils import get_tokenizer
from collections import Counter
import random
from torch.utils.data import Subset, DataLoader
import time

训练

# 主函数
if __name__ == '__main__':
    start_time = time.time()  # 开始计时

    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

    #terminology = load_terminology_dictionary('../dataset/en-zh.dic')
    terminology = load_terminology_dictionary('./dataset/en-zh.dic')

    # 加载数据
    dataset = TranslationDataset('./dataset/train.txt',terminology = terminology)
    # 选择数据集的前N个样本进行训练
    N = 1000  #int(len(dataset) * 1)  # 或者你可以设置为数据集大小的一定比例，如 int(len(dataset) * 0.1)
    subset_indices = list(range(N))
    subset_dataset = Subset(dataset, subset_indices)
    train_loader = DataLoader(subset_dataset, batch_size=32, shuffle=True, collate_fn=collate_fn)

    # 定义模型参数
    INPUT_DIM = len(dataset.en_vocab)
    OUTPUT_DIM = len(dataset.zh_vocab)
    ENC_EMB_DIM = 256
    DEC_EMB_DIM = 256
    HID_DIM = 512
    N_LAYERS = 2
    ENC_DROPOUT = 0.5
    DEC_DROPOUT = 0.5

    # 初始化模型
    enc = Encoder(INPUT_DIM, ENC_EMB_DIM, HID_DIM, N_LAYERS, ENC_DROPOUT)
    dec = Decoder(OUTPUT_DIM, DEC_EMB_DIM, HID_DIM, N_LAYERS, DEC_DROPOUT)
    model = Seq2Seq(enc, dec, device).to(device)

    # 定义优化器和损失函数
    optimizer = optim.Adam(model.parameters())
    criterion = nn.CrossEntropyLoss(ignore_index=dataset.zh_word2idx['<pad>'])

    # 训练模型
    N_EPOCHS = 10
    CLIP = 1

    for epoch in range(N_EPOCHS):
        train_loss = train(model, train_loader, optimizer, criterion, CLIP)
        print(f'Epoch: {epoch+1:02} | Train Loss: {train_loss:.3f}')
        
    # 在训练循环结束后保存模型
    torch.save(model.state_dict(), 'translation_model_GRU.pth')
    
    end_time = time.time()  # 结束计时

    # 计算并打印运行时间
    elapsed_time_minute = (end_time - start_time)/60
    print(f"Total running time: {elapsed_time_minute:.2f} minutes")

输出结果如下：

Epoch: 01 | Train Loss: 6.555
Epoch: 02 | Train Loss: 6.060
Epoch: 03 | Train Loss: 6.030
Epoch: 04 | Train Loss: 5.988
Epoch: 05 | Train Loss: 5.922
Epoch: 06 | Train Loss: 5.868
Epoch: 07 | Train Loss: 5.799
Epoch: 08 | Train Loss: 5.700
Epoch: 09 | Train Loss: 5.603
Epoch: 10 | Train Loss: 5.518
Total running time: 1.31 minutes

在测试集上进行推理

# 主函数
if __name__ == '__main__':
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

    # 加载术语词典
    terminology = load_terminology_dictionary('./dataset/en-zh.dic')
    # 加载数据集和模型
    dataset = TranslationDataset('./dataset/train.txt',terminology = terminology)

    # 定义模型参数
    INPUT_DIM = len(dataset.en_vocab)
    OUTPUT_DIM = len(dataset.zh_vocab)
    ENC_EMB_DIM = 256
    DEC_EMB_DIM = 256
    HID_DIM = 512
    N_LAYERS = 2
    ENC_DROPOUT = 0.5
    DEC_DROPOUT = 0.5

    # 初始化模型
    enc = Encoder(INPUT_DIM, ENC_EMB_DIM, HID_DIM, N_LAYERS, ENC_DROPOUT)
    dec = Decoder(OUTPUT_DIM, DEC_EMB_DIM, HID_DIM, N_LAYERS, DEC_DROPOUT)
    model = Seq2Seq(enc, dec, device).to(device)

    # 加载训练好的模型
    model.load_state_dict(torch.load('translation_model_GRU.pth'))
    
    save_dir = './dataset/submit.txt'
    inference(model, dataset, src_file="./dataset/test_en.txt", save_dir = save_dir, terminology = terminology, device = device)
    print(f"翻译完成！文件已保存到{save_dir}")

输出结果如下：

翻译完成！文件已保存到./dataset/submit.txt

上传至Kaggle