赛题链接:https://challenge.xfyun.cn/topic/info?type=machine-translation-2024
赛题解读
安装库
spacy
1.查看本地spacy版本
pip show spacy
我安装3.6.0
pip install en_core_web_sm-3.6.0.tar.gz
en_core_web_sm下载链接:https://github.com/explosion/spacy-models/releases?q=en_core_web_sm&expanded=true
数据预处理
赛题数据
- 训练集:双语数据 - 中英14万余双语句对
- 开发集:英中1000双语句对
- 测试集:英中1000双语句对
- 术语词典:英中2226条
# 定义数据集类
# 修改TranslationDataset类以处理术语
class TranslationDataset(Dataset):
def __init__(self, filename, terminology):
self.data = []
with open(filename, 'r', encoding='utf-8') as f:
for line in f:
en, zh = line.strip().split('\t')
self.data.append((en, zh))
self.terminology = terminology
# 创建词汇表,注意这里需要确保术语词典中的词也被包含在词汇表中
self.en_tokenizer = get_tokenizer('basic_english')
self.zh_tokenizer = list # 使用字符级分词
en_vocab = Counter(self.terminology.keys()) # 确保术语在词汇表中
zh_vocab = Counter()
for en, zh in self.data:
en_vocab.update(self.en_tokenizer(en))
zh_vocab.update(self.zh_tokenizer(zh))
# 添加术语到词汇表
self.en_vocab = ['<pad>', '<sos>', '<eos>'] + list(self.terminology.keys()) + [word for word, _ in en_vocab.most_common(10000)]
self.zh_vocab = ['<pad>', '<sos>', '<eos>'] + [word for word, _ in zh_vocab.most_common(10000)]
self.en_word2idx = {word: idx for idx, word in enumerate(self.en_vocab)}
self.zh_word2idx = {word: idx for idx, word in enumerate(self.zh_vocab)}
def __len__(self):
return len(self.data)
def __getitem__(self, idx):
en, zh = self.data[idx]
en_tensor = torch.tensor([self.en_word2idx.get(word, self.en_word2idx['<sos>']) for word in self.en_tokenizer(en)] + [self.en_word2idx['<eos>']])
zh_tensor = torch.tensor([self.zh_word2idx.get(word, self.zh_word2idx['<sos>']) for word in self.zh_tokenizer(zh)] + [self.zh_word2idx['<eos>']])
return en_tensor, zh_tensor
def collate_fn(batch):
en_batch, zh_batch = [], []
for en_item, zh_item in batch:
en_batch.append(en_item)
zh_batch.append(zh_item)
# 对英文和中文序列分别进行填充
en_batch = nn.utils.rnn.pad_sequence(en_batch, padding_value=0, batch_first=True)
zh_batch = nn.utils.rnn.pad_sequence(zh_batch, padding_value=0, batch_first=True)
return en_batch, zh_batch