summary-reward-no-reference/step1_encode_doc_summ.py at master · yg211/summary-reward-no-reference · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
import numpy as np
from nltk import PorterStemmer
from nltk.corpus import stopwords
from tqdm import tqdm
import pickle
import torch
from pytorch_transformers import *

from scorer.data_helper.json_reader import read_sorted_scores, read_articles, read_processed_scores, read_scores
from helpers.data_helpers import sent2stokens_wostop, sent2tokens_wostop, sent2stokens, text_normalization


def raw_bert_encoder(model, tokenizer, sent_list, stride=128, gpu=True):
    merged_text = ''
    for ss in sent_list: merged_text += ss+' '
    tokens = tokenizer.encode(merged_text)

    model.eval()
    with torch.no_grad():
        if len(tokens) <= 510:
            tokens = torch.tensor(tokens).unsqueeze(0)
            if gpu:
                tokens = tokens.to('cuda')
                model.to('cuda')
            vv = model(tokens)[0][0].data.cpu().numpy()
            vv = np.mean(vv,axis=0)
        else:
            end_pointer = stride
            batch = []
            real_length = []
            att_masks = []
            while True:
                start_pointer = end_pointer-510
                if start_pointer < 0: start_pointer = 0
                if start_pointer >= len(tokens): break
                if end_pointer <= len(tokens):
                    batch.append(tokens[start_pointer:end_pointer])
                    real_length.append(end_pointer-start_pointer)
                    att_masks.append([1]*real_length[-1])
                else:
                    batch.append(tokens[start_pointer:end_pointer])
                    real_length.append(len(tokens)-start_pointer)
                    att_masks.append([1] * real_length[-1])
                end_pointer += stride
                #print(len(batch[-1]))

            #padding
            longest = max(real_length)
            for ii in range(len(batch)):
                batch[ii] += [0] * (longest-real_length[ii])
                att_masks[ii] += [0] * (longest-real_length[ii])

            batch = torch.tensor(batch)
            att_masks = torch.tensor(att_masks)
            if gpu:
                batch = batch.to('cuda')
                att_masks = att_masks.to('cuda')
                model.to('cuda')

            last_layers = model(input_ids=batch,attention_mask=att_masks)[0].data.cpu().numpy()
            vectors = []
            for ii,bb in enumerate(last_layers):
                vectors.append(np.mean(bb[:real_length[ii]],axis=0))
            vv = np.mean(vectors,axis=0)

    return vv


def encode_doc_summ(stem=False, remove_stop=False):
    bert_tokenizer = BertTokenizer.from_pretrained('bert-large-uncased')
    bert_model = BertModel.from_pretrained('bert-large-uncased')

    sorted_scores = read_sorted_scores()
    input_articles, _ = read_articles()

    stopwords_list = set(stopwords.words("english"))
    stemmer = PorterStemmer()
    vec_dic = {}

    for i, (article_id, scores_list) in tqdm(enumerate(sorted_scores.items())):
        vec_dic[article_id] = {}
        article = [entry['article'] for entry in input_articles if entry['id']==article_id][0]
        ref_summ = scores_list[0]['ref']

        if stem and remove_stop:
            sys_summs = [" ".join(sent2stokens_wostop(s['sys_summ'], stemmer, stopwords_list, 'english', True)) for s in scores_list]
            ref_summ = " ".join(sent2stokens_wostop(ref_summ, stemmer, stopwords_list, 'english', True))
            article = " ".join(sent2stokens_wostop(article, stemmer, stopwords_list, 'english', True))
        elif not stem and remove_stop:
            sys_summs = [" ".join(sent2tokens_wostop(s['sys_summ'], stopwords_list, 'english', True)) for s in scores_list]
            ref_summ = " ".join(sent2tokens_wostop(ref_summ, stopwords_list, 'english', True))
            article = " ".join(sent2tokens_wostop(article, stopwords_list, 'english', True))
        elif not remove_stop and stem:
            sys_summs = [" ".join(sent2stokens(s['sys_summ'], stemmer, 'english', True)) for s in
                         scores_list]
            ref_summ = " ".join(sent2stokens(ref_summ, stemmer, 'english', True))
            article = " ".join(sent2stokens(article, stemmer, 'english', True))
        else:
            sys_summs = [s['sys_summ'] for s in scores_list]

        summ_ids = [s['summ_id'] for s in scores_list]

        # clean text
        sys_summs = [text_normalization(s) for s in sys_summs]
        ref_summ = text_normalization(ref_summ)
        article = text_normalization(article)

        vec_dic[article_id]['article'] = raw_bert_encoder(bert_model, bert_tokenizer, [article])
        vec_dic[article_id]['ref'] = raw_bert_encoder(bert_model, bert_tokenizer, [ref_summ])
        for i,sid in enumerate(summ_ids):
            vec_dic[article_id]['sys_summ{}'.format(sid)] = raw_bert_encoder(bert_model, bert_tokenizer, [sys_summs[i]])

    save_file_name = 'doc_summ_bert_vectors'
    if stem: save_file_name+'_stem'
    if remove_stop: save_file_name+'_removeStop'
    save_file_name += '.pkl'
    with open('data/'+save_file_name,'wb') as ff:
        pickle.dump(vec_dic,ff)


if __name__ == '__main__':
    encode_doc_summ(stem=False,remove_stop=False)