69用于预训练BERT的数据集

点击查看代码

import os
import random
import torch
from d2l import torch as d2l

#@save
d2l.DATA_HUB['wikitext-2'] = (
    'https://s3.amazonaws.com/research.metamind.io/wikitext/'
    'wikitext-2-v1.zip', '3c914d17d80b1459be871a5039ac23e752a53cbe')

#@save
def _read_wiki(data_dir):
    file_name = os.path.join(data_dir, 'wiki.train.tokens')
    with open(file_name, 'r', encoding = 'utf-8') as f:
        lines = f.readlines()
    # 大写字母转换为小写字母
    paragraphs = [line.strip().lower().split(' . ')
                  for line in lines if len(line.split(' . ')) >= 2]
    random.shuffle(paragraphs)
    return paragraphs

# 生成下一句预测任务的数据
#@save
def _get_next_sentence(sentence, next_sentence, paragraphs):
    # 预测一个句子对中两个句子是不是相邻
    # 50%概率选择相邻句子对，50%概率选择随机句子对
    # print('type(paragraphs) : ', type(paragraphs))
    # print('len(paragraphs) : ', len(paragraphs))
    # print('paragraphs : ', paragraphs)
    # print('random.choice(paragraphs) : ', random.choice(paragraphs))
    # print('random.choice(random.choice(paragraphs)) : ', random.choice(random.choice(paragraphs)))
    if random.random() < 0.5:
        is_next = True
    else:
        # paragraphs是三重列表的嵌套
        next_sentence = random.choice(random.choice(paragraphs))
        is_next = False
    return sentence, next_sentence, is_next

# 下一个句子预测
#@save
def _get_nsp_data_from_paragraph(paragraph, paragraphs, vocab, max_len):
    nsp_data_from_paragraph = []
    for i in range(len(paragraph) - 1):
        tokens_a, tokens_b, is_next = _get_next_sentence(paragraph[i], paragraph[i + 1], paragraphs)
        # 考虑1个'<cls>'词元和2个'<sep>'词元
        # 只会终止执行本次循环中剩下的代码，直接从下一次循环继续执行。
        if len(tokens_a) + len(tokens_b) + 3 > max_len:
            continue
        # 转换为BERT输入
        tokens, segments = d2l.get_tokens_and_segments(tokens_a, tokens_b)
        # 生产下一个句子预测数据
        nsp_data_from_paragraph.append((tokens, segments, is_next))
    return nsp_data_from_paragraph


# 生成遮蔽语言模型任务的数据
# 从BERT输入序列生成遮蔽语言模型的训练样本
#@save
def _replace_mlm_tokens(tokens, candidate_pred_positions, num_mlm_preds, vocab):
    """
    :param tokens: BERT输入序列的词元的列表
    :param candidate_pred_positions:不包括特殊词元的BERT输入序列的词元索引的列表（特殊词元在遮蔽语言模型任务中不被预测）
    :param num_mlm_preds:预测的数量（选择15%要预测的随机词元）
    :param vocab:
    :return:替换后的输入词元、发生预测的词元索引和这些预测的标签
    """
    # 为遮蔽语言模型的输入创建新的词元副本，其中输入可能包含替换的“<mask>”或随机词元
    mlm_input_tokens = [token for token in tokens]
    pred_positions_and_labels = []
    # 打乱后用于在遮蔽语言模型任务中获取15%的随机词元进行预测
    random.shuffle(candidate_pred_positions)
    for mlm_pred_position in candidate_pred_positions:
        if len(pred_positions_and_labels) >= num_mlm_preds:
            break
        masked_token = None
        # 80%的概率：将词替换为“<mask>”词元
        if random.random() < 0.8:
            masked_token = '<mask>'
        else:
            # 10%的概率：保持词不变
            if random.random() < 0.5:
                masked_token = tokens[mlm_pred_position]
            # 10%的概率：用随机词替换该词
            else:
                masked_token = random.choice(vocab.idx_to_token)
        mlm_input_tokens[mlm_pred_position] = masked_token
        pred_positions_and_labels.append((mlm_pred_position, tokens[mlm_pred_position]))
    return mlm_input_tokens, pred_positions_and_labels

# 将BERT输入序列（tokens）作为输入，并返回输入词元的索引、发生预测的词元索引以及这些预测的标签索引。
#@save
def _get_mlm_data_from_tokens(tokens, vocab):
    candidate_pred_positions = []
    # tokens是一个字符串列表
    for i, token in enumerate(tokens):
        # 在遮蔽语言模型任务中不会预测特殊词元
        if token in ['<cls>', '<sep>']:
            continue
        candidate_pred_positions.append(i)
    # 遮蔽语言模型任务中预测15%的随机词元
    num_mlm_preds = max(1, round(len(tokens) * 0.15))
    mlm_input_tokens, pred_positions_and_labels = _replace_mlm_tokens(
        tokens, candidate_pred_positions, num_mlm_preds, vocab)
    # pred_positions_and_labels.append((mlm_pred_position, tokens[mlm_pred_position]))
    pred_positions_and_labels = sorted(pred_positions_and_labels,key=lambda x: x[0])
    pred_positions = [v[0] for v in pred_positions_and_labels]
    mlm_pred_labels = [v[1] for v in pred_positions_and_labels]
    # vocab[mlm_input_tokens] - > mask
    return vocab[mlm_input_tokens], pred_positions, vocab[mlm_pred_labels]


# 将文本转换为预训练数据集
#@save
def _pad_bert_inputs(examples, max_len, vocab):
    max_num_mlm_preds = round(max_len * 0.15)
    all_token_ids, all_segments, valid_lens,  = [], [], []
    # 拼接 all_mlm_weights = 0 否则为1
    all_pred_positions, all_mlm_weights, all_mlm_labels = [], [], []
    nsp_labels = []
    # _get_nsp_data_from_paragraph
    # return nsp_data_from_paragraph  -- (tokens, segments, is_next)
    # _get_mlm_data_from_tokens
    # return vocab[mlm_input_tokens], pred_positions, vocab[mlm_pred_labels]
    for (token_ids, pred_positions, mlm_pred_label_ids, segments, is_next) in examples:
        all_token_ids.append(torch.tensor(token_ids + [vocab['<pad>']] * (max_len - len(token_ids)), dtype=torch.long))
        all_segments.append(torch.tensor(segments + [0] * (max_len - len(segments)), dtype=torch.long))
        # valid_lens不包括'<pad>'的计数
        valid_lens.append(torch.tensor(len(token_ids), dtype=torch.float32))
        all_pred_positions.append(torch.tensor(pred_positions + [0] * (max_num_mlm_preds - len(pred_positions)), dtype=torch.long))
        # 填充词元的预测将通过乘以0权重在损失中过滤掉
        all_mlm_weights.append(torch.tensor([1.0] * len(mlm_pred_label_ids) + [0.0] * (max_num_mlm_preds - len(pred_positions)),
                                dtype=torch.float32))
        all_mlm_labels.append(torch.tensor(mlm_pred_label_ids + [0] * (max_num_mlm_preds - len(mlm_pred_label_ids)), dtype=torch.long))
        nsp_labels.append(torch.tensor(is_next, dtype=torch.long))
    return (all_token_ids, all_segments, valid_lens, all_pred_positions, all_mlm_weights, all_mlm_labels, nsp_labels)

#@save
class _WikiTextDataset(torch.utils.data.Dataset):
    def __init__(self, paragraphs, max_len):
        #   输入paragraphs[i]是代表段落的句子字符串列表；
        # 而输出paragraphs[i]是代表段落的句子列表，其中每个句子都是词元列表
        paragraphs = [d2l.tokenize(paragraph, token='word') for paragraph in paragraphs]
        sentences = [sentence for paragraph in paragraphs for sentence in paragraph]
        self.vocab = d2l.Vocab(sentences, min_freq=5, reserved_tokens=['<pad>', '<mask>', '<cls>', '<sep>'])
        # 获取下一句子预测任务的数据
        examples = []
        for paragraph in paragraphs:
            # def _get_nsp_data_from_paragraph(paragraph, paragraphs, vocab, max_len):
            #       nsp_data_from_paragraph.append((tokens, segments, is_next))
            #   return nsp_data_from_paragraph
            # extend() 函数用于在列表末尾一次性追加另一个序列中的多个值（用新列表扩展原来的列表）。
            examples.extend(_get_nsp_data_from_paragraph(paragraph, paragraphs, self.vocab, max_len))
        # 获取遮蔽语言模型任务的数据
        # def _get_mlm_data_from_tokens(tokens, vocab):
        #   return vocab[mlm_input_tokens], pred_positions, vocab[mlm_pred_labels]
        examples = [(_get_mlm_data_from_tokens(tokens, self.vocab) + (segments, is_next)) for tokens, segments, is_next in examples]
        # 填充输入
        # for (token_ids, pred_positions, mlm_pred_label_ids, segments, is_next) in examples:
        (self.all_token_ids, self.all_segments, self.valid_lens, self.all_pred_positions, self.all_mlm_weights,
         self.all_mlm_labels, self.nsp_labels) = _pad_bert_inputs(examples, max_len, self.vocab)

    def __getitem__(self, idx):
        return (self.all_token_ids[idx], self.all_segments[idx],
                self.valid_lens[idx], self.all_pred_positions[idx],
                self.all_mlm_weights[idx], self.all_mlm_labels[idx],
                self.nsp_labels[idx])

    def __len__(self):
        return len(self.all_token_ids)

#@save
def load_data_wiki(batch_size, max_len):
    """加载WikiText-2数据集"""
    num_workers = 0
    data_dir = d2l.download_extract('wikitext-2', 'wikitext-2')
    paragraphs = _read_wiki(data_dir)
    train_set = _WikiTextDataset(paragraphs, max_len)
    train_iter = torch.utils.data.DataLoader(train_set, batch_size, shuffle=True, num_workers=num_workers)
    return train_iter, train_set.vocab

batch_size, max_len = 512, 64
train_iter, vocab = load_data_wiki(batch_size, max_len)

# tokens_X         torch.Size([512, 64])
# segments_X       torch.Size([512, 64])
# valid_lens_x     torch.Size([512])
# pred_positions_X torch.Size([512, 10]) 预测多少个位置 64*0.15
# mlm_weights_X    torch.Size([512, 10])
# mlm_Y            torch.Size([512, 10])
# nsp_y            torch.Size([512])
for (tokens_X, segments_X, valid_lens_x, pred_positions_X, mlm_weights_X, mlm_Y, nsp_y) in train_iter:
    print(tokens_X.shape, segments_X.shape, valid_lens_x.shape, pred_positions_X.shape, mlm_weights_X.shape, mlm_Y.shape, nsp_y.shape)
    break