DASK/data_loader.py at main · hikaru-nara/DASK · 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
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
import os, numpy as np
import scipy.sparse as sp
from gensim.corpora import Dictionary as gensim_dico
from brain.knowgraph import KnowledgeGraph


class bdek_train_dataset(torch.utils.data.IterableDataset):
    def __init__(self, source_name, target_name, graph_path max_words=5000, supervision_rate=0.1, repeat=1):
        super(bdek_dataset, self).__init__()
        X_s, Y_s, d_s, g_s, self.X_tu, self.Y_tu, self.d_tu, self.g_tu, X_semi, Y_semi, d_semi, g_semi, \
            _,_,_,_ = forge_dataset(source_name, target_name, max_words, supervision_rate)
        # print(X_s.shape)
        self.X_l = np.concatenate((X_s, X_semi))
        self.Y_l = np.concatenate((Y_s, Y_semi))
        self.d_l = np.concatenate((d_s, d_semi))
        self.g_l = np.concatenate((g_s, g_semi))
        self.kg = KnowledgeGraph(graph_path, predicate)

def create_vocab(sentence_list):
    for sent in sentence_list:
        for word in

def forge_dataset(source_name, target_name, max_words, supervision_rate):
    t_s, y_s = get_labeled_dataset(source_name)
    d_s = np.zeros(len(t_s))
    t_tu = get_unlabeled_dataset(target_name)
    d_tu = np.ones(len(t_tu))
    t_tl, y_tl = get_labeled_dataset(target_name)
    d_tl = np.ones(len(t_tl))
    g_s, g_tu, g_tl = get_graph_feature(source_name, target_name)
    vocab = create_vocab(t_s+t_tu+t_tl)


def get_graph_feature(source_name, target_name):
    '''
    get the graph features from source and target dataset
    '''
    g_s = np.load(open('graph_features/sf_' + d1 +'_small_5000.np', 'rb'), allow_pickle=True)
    g_tu = np.load(open('graph_features/sf_' + d2 + '_small_5000.np', 'rb'), allow_pickle=True)
    g_tl = np.load(open('graph_features/sf_'+ d2 + '_test_5000.np', 'rb'), allow_pickle=True)
    return g_s, g_tu, g_tl


###########################


def parse_processed_amazon_dataset(FNames, max_words):
    datasets = {}
    dico = gensim_dico()

    # First pass on document to build dictionary
    for fname in FNames:
        f = open(fname)
        for l in f:
            tokens = l.split(sep=' ')
            label_string = tokens[-1]
            tokens_list=[]
            for tok in tokens[:-1]:
                ts, tfreq = tok.split(':')
                freq = int(tfreq)
                tokens_list += [ts]*freq

            _ = dico.doc2bow(tokens_list, allow_update=True)

        f.close()

    # Preprocessing_options
    dico.filter_extremes(no_below=2, keep_n=max_words)
    dico.compactify()

    for fname in FNames:
        X = []
        Y = []
        docid = -1
        f = open(fname)
        for l in f:
            tokens = l.split(sep=' ')
            label_string = tokens[-1]
            tokens_list = []
            for tok in tokens[:-1]:
                ts, tfreq = tok.split(':')
                freq = int(tfreq)
                tokens_list += [ts]*freq

            count_list = dico.doc2bow(tokens_list, allow_update=False)

            docid += 1

            X.append((docid, count_list))

            # Preprocess Label
            ls, lvalue = label_string.split(':')
            if ls == "#label#":
                if lvalue.rstrip() == 'positive':
                    lv = 1
                    Y.append(lv)
                elif lvalue.rstrip() == 'negative':
                    lv = 0
                    Y.append(lv)
                else:
                    raise Exception("Invalid Label Value")
            else:
                raise Exception('Invalid Format')

        datasets[fname] = (X, np.array(Y))
        f.close()
        del f

    return datasets, dico


def count_list_to_sparse_matrix(X_list, dico):
    ndocs = len(X_list)
    voc_size = len(dico.keys())


    X_spmatrix = sp.lil_matrix((ndocs, voc_size))
    for did, counts in X_list:
        for wid, freq in counts:
            X_spmatrix[did, wid]=freq

    return X_spmatrix.tocsr()


def get_dataset_path(domain_name, exp_type):
    prefix ='./dataset/'
    if exp_type == 'small':
        fname = 'labelled.review'
    elif exp_type == 'all':
        fname = 'all.review'
    elif exp_type == 'test':
        fname = 'unlabeled.review'

    return os.path.join(prefix, domain_name, fname)

def split_semi_supervision(texts, labels, supervision_rate=0.1):

    return (None, None, None, None)


def get_dataset(source_name, target_name, max_words=5000):

    source_path  = get_dataset_path(source_name, 'small')
    target_path1 = get_dataset_path(target_name, 'small')
    target_path2 = get_dataset_path(target_name, 'test')

    dataset_list = [source_path, target_path1, target_path2]
    datasets, dico = parse_processed_amazon_dataset(dataset_list, max_words, supervision_rate)

    L_s, Y_s = datasets[source_path]
    L_tu, Y_tu = datasets[target_path1]
    L_tl, Y_tl = datasets[target_path2]

    # L_semi, Y_semi, L_test, Y_test = split_semi_supervision(L_t2, Y_t2)

    X_s  = count_list_to_sparse_matrix(L_s,  dico)
    X_tu = count_list_to_sparse_matrix(L_tu, dico)
    X_tl = count_list_to_sparse_matrix(L_tl, dico)

    d_s = np.zeros(len(L_s)).astype(np.int64)
    d_tu = np.ones(len(L_tu)).astype(np.int64)
    d_tl = np.ones(len(L_tl)).astype(np.int64)

    return X_s, Y_s, d_s, X_tu, Y_tu, d_tu, X_tl, Y_tl, d_tl, dico

def get_graph_feature(source_name, target_name):
    '''
    get the graph features from source and target dataset
    '''
    g_s = np.load(open('graph_features/sf_' + d1 +'_small_5000.np', 'rb'), allow_pickle=True)
    g_tu = np.load(open('graph_features/sf_' + d2 + '_small_5000.np', 'rb'), allow_pickle=True)
    g_tl = np.load(open('graph_features/sf_'+ d2 + '_test_5000.np', 'rb'), allow_pickle=True)
    return g_s, g_tu, g_tl

def get_commonsense_graph(graph_path, predicate=True):
    return KnowledgeGraph(graph_path, predicate)

def forge_dataset(source_name, target_name, max_words=5000, supervision_rate=0.1):
    g_s, g_tu, g_tl = get_graph_feature(source_name, target_name)
    X_s, Y_s, d_s, X_tu, Y_tu, d_tu, X_tl, Y_tl, d_tl, _ = get_dataset(source_name, target_name, max_words)
    X_semi, Y_semi, d_semi, g_semi, X_te, Y_te, d_te, g_te = split_semi_supervision(X_tl, Y_tl, d_tl, g_tl)
    return X_s, Y_s, d_s, g_s, X_tu, Y_tu, d_tu, g_tu, X_semi, Y_semi, d_semi, g_semi, X_te, Y_te, d_te, g_te

if __name__=='__main__':
    D_tr = bdek_train_dataset('books','dvd')