데이터 사전 처리

'''
  code by Tae Hwan Jung(Jeff Jung) @graykode
  Reference : https://github.com/jadore801120/attention-is-all-you-need-pytorch
              https://github.com/JayParks/transformer, https://github.com/dhlee347/pytorchic-bert
'''
import math
import re
from random import *
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.autograd import Variable
import pdb

# BERT Parameters
maxlen = 30
batch_size = 6
max_pred = 5 # max tokens of prediction
n_layers = 6
n_heads = 12
d_model = 768
d_ff = 768*4 # 4*d_model, FeedForward dimension
d_k = d_v = 64  # dimension of K(=Q), V
n_segments = 2

text = (
    'Hello, how are you? I am Romeo.
'
    'Hello, Romeo My name is Juliet. Nice to meet you.
'
    'Nice meet you too. How are you today?
'
    'Great. My baseball team won the competition.
'
    'Oh Congratulations, Juliet
'
    'Thanks you Romeo'
)
sentences = re.sub("[.,!?\\-]", '', text.lower()).split('
') # filter '.', ',', '?', '!'
word_list = list(set(" ".join(sentences).split()))
word_dict = {'[PAD]' : 0, '[CLS]' : 1, '[SEP]' : 2, '[MASK]' : 3}
for i, w in enumerate(word_list):
    word_dict[w] = i + 4
number_dict = {i: w for i, w in enumerate(word_dict)}
vocab_size = len(word_dict)

token_list = list()
for sentence in sentences:
    arr = [word_dict[s] for s in sentence.split()]
    token_list.append(arr)

Batch 생성

# sample IsNext and NotNext to be same in small batch size
def make_batch():
    batch = []
    positive = negative = 0
    while positive != batch_size/2 or negative != batch_size/2:
        tokens_a_index, tokens_b_index= randrange(len(sentences)), randrange(len(sentences)) # sample random index in sentences
        tokens_a, tokens_b= token_list[tokens_a_index], token_list[tokens_b_index]
        input_ids = [word_dict['[CLS]']] + tokens_a + [word_dict['[SEP]']] + tokens_b + [word_dict['[SEP]']]
        segment_ids = [0] * (1 + len(tokens_a) + 1) + [1] * (len(tokens_b) + 1)

        # MASK LM
        n_pred =  min(max_pred, max(1, int(round(len(input_ids) * 0.15)))) # 15 % of tokens in one sentence
        #mask   ，15%   ，          
        cand_maked_pos = [i for i, token in enumerate(input_ids)
                          if token != word_dict['[CLS]'] and token != word_dict['[SEP]']]
        shuffle(cand_maked_pos)
        masked_tokens, masked_pos = [], []
        for pos in cand_maked_pos[:n_pred]:
            masked_pos.append(pos)
            masked_tokens.append(input_ids[pos])
            if random() < 0.8:  # 80%
                input_ids[pos] = word_dict['[MASK]'] # make mask
            elif random() < 0.5:  # 10%
                index = randint(0, vocab_size - 1) # random index in vocabulary
                input_ids[pos] = word_dict[number_dict[index]] # replace

        # Zero Paddings
        pdb.set_trace()
        n_pad = maxlen - len(input_ids)
        input_ids.extend([0] * n_pad)
        segment_ids.extend([0] * n_pad)

        # Zero Padding (100% - 15%) tokens
        if max_pred > n_pred:
            n_pad = max_pred - n_pred
            masked_tokens.extend([0] * n_pad)
            masked_pos.extend([0] * n_pad)

        if tokens_a_index + 1 == tokens_b_index and positive < batch_size/2:
            batch.append([input_ids, segment_ids, masked_tokens, masked_pos, True]) # IsNext
            positive += 1
        elif tokens_a_index + 1 != tokens_b_index and negative < batch_size/2:
            batch.append([input_ids, segment_ids, masked_tokens, masked_pos, False]) # NotNext
            negative += 1
    return batch
# Proprecessing Finished