PCNN_TensorFlow/context_split.py at master · NekoApocalypse/PCNN_TensorFlow · 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
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
import numpy as np


def mask_gen(origin, pos1, pos2):
    mask_left = [0] * len(origin)
    mask_mid = [0] * len(origin)
    mask_right = [0] * len(origin)
    mask_left[:pos1 + 1] = [1] * (pos1 + 1)
    mask_mid[pos1: pos2 + 1] = [1] * (pos2 + 1 - pos1)
    mask_right[pos2:] = [1] * (len(origin) - pos2)
    assert len(origin) == len(mask_left)
    assert len(origin) == len(mask_mid)
    assert len(origin) == len(mask_right)
    return mask_left, mask_mid, mask_left


def sentence_pad(origin, pos1, pos2, pad_token):
    part_left = [pad_token] * len(origin)
    part_mid = [pad_token] * len(origin)
    part_right = [pad_token] * len(origin)

    part_left[:pos1 + 1] = origin[:pos1 + 1]
    part_mid[pos1: pos2 + 1] = origin[pos1: pos2 + 1]
    part_right[pos2:] = origin[pos2:]
    return part_left, part_mid, part_right


def context_mask(train_word, train_pos1, train_pos2, train_y):
    skipped_sentences = 0
    skipped_entity_pair = 0
    mask_left = []
    mask_mid = []
    mask_right = []
    context_x = []
    context_y = []
    context_pos1 = []
    context_pos2 = []
    for i in range(len(train_word)):
        buffer_left = []
        buffer_mid = []
        buffer_right = []
        buffer_x = []
        buffer_pos1 = []
        buffer_pos2 = []
        for j in range(len(train_word[i])):
            if not ((61 in train_pos1[i][j]) and (61 in train_pos2[i][j])):
                # print(i, j)
                skipped_sentences += 1
                continue
            # Warning:
            # training data contains python lists at the last dimension,
            # but testing data contains np.ndarray at the last dimension.
            if isinstance(train_pos1[i][j], np.ndarray):
                pos1_id = train_pos1[i][j].tolist().index(61)
                pos2_id = train_pos2[i][j].tolist().index(61)
            else:
                pos1_id = train_pos1[i][j].index(61)
                pos2_id = train_pos2[i][j].index(61)

            tmp_left, tmp_mid, tmp_right = mask_gen(
                train_word[i][j], pos1_id, pos2_id)
            buffer_left.append(tmp_left)
            buffer_mid.append(tmp_mid)
            buffer_right.append(tmp_right)
            buffer_x.append(train_word[i][j])
            buffer_pos1.append(train_pos1[i][j])
            buffer_pos2.append(train_pos2[i][j])

        if not buffer_left:
            skipped_entity_pair += 1
            continue

        assert len(buffer_left) == len(buffer_x)
        mask_left.append(buffer_left)
        mask_mid.append(buffer_mid)
        mask_right.append(buffer_right)
        context_x.append(buffer_x)
        context_pos1.append(buffer_pos1)
        context_pos2.append(buffer_pos2)
        context_y.append(train_y[i])

    assert len(mask_left) == len(context_x)
    print('Skipped sentences: ', skipped_sentences)
    print('Skipped entity pairs: ', skipped_entity_pair)
    return mask_left, mask_mid, mask_right, context_x, context_pos1, context_pos2, context_y


def context_split(train_word, train_pos1, train_pos2, train_y):
    # [num_position, num_entity, num_sentences, sentence_len]
    # special_tokens: 'BLANK', 'UNK'
    special_token = np.load('./data/special_token.npy')
    blank_token = special_token[0]

    context_word = [[], [], []]
    context_pos1 = [[], [], []]
    context_pos2 = [[], [], []]
    context_y = []
    skipped_sentences = 0
    skipped_entity_pair = 0
    input_length = len(train_word[0][0])
    for i in range(len(train_word)):
        # each buffer contains three list of sentences.
        word_buffer = [[], [], []]
        pos1_buffer = [[], [], []]
        pos2_buffer = [[], [], []]
        for j in range(len(train_word[i])):
            if not ((61 in train_pos1[i][j]) and (61 in train_pos2[i][j])):
                # print(i, j)
                skipped_sentences += 1
                continue

            # Warning:
            # training data contains python lists at the last dimension,
            # but testing data contains np.ndarray at the last dimension.
            # Note:
            # The output of sentence_pad should all be lists.
            if isinstance(train_pos1[i][j], np.ndarray):
                pos1_id = train_pos1[i][j].tolist().index(61)
                pos2_id = train_pos2[i][j].tolist().index(61)
            else:
                pos1_id = train_pos1[i][j].index(61)
                pos2_id = train_pos2[i][j].index(61)
            # each part is padded to input_length
            word_left, word_mid, word_right = \
                sentence_pad(train_word[i][j], pos1_id, pos2_id, blank_token)
            word_buffer[0].append(word_left)
            word_buffer[1].append(word_mid)
            word_buffer[2].append(word_right)

            pos1_left, pos1_mid, pos1_right = \
                sentence_pad(train_pos1[i][j], pos1_id, pos2_id, blank_token)
            pos1_buffer[0].append(pos1_left)
            pos1_buffer[1].append(pos1_mid)
            pos1_buffer[2].append(pos1_right)

            pos2_left, pos2_mid, pos2_right = \
                sentence_pad(train_pos2[i][j], pos1_id, pos2_id, blank_token)
            pos2_buffer[0].append(pos2_left)
            pos2_buffer[1].append(pos2_mid)
            pos2_buffer[2].append(pos2_right)

        if not word_buffer[0]:
            skipped_entity_pair += 1
            # print(i)
            continue

        for t in range(3):
            context_word[t].append(word_buffer[t])
            context_pos1[t].append(pos1_buffer[t])
            context_pos2[t].append(pos2_buffer[t])
        context_y.append(train_y[i])
    print('Skipped sentences: ', skipped_sentences)
    print('Skipped entity pairs: ', skipped_entity_pair)
    return context_word, context_pos1, context_pos2, context_y