-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataset.py
245 lines (227 loc) · 8.58 KB
/
dataset.py
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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
from gensim.models import Word2Vec
from torch.utils import data
from tqdm import tqdm
import numpy as np
from typing import List
import torch
from collections import Counter
from sklearn_extra.cluster import KMedoids
class TrainDataset(data.Dataset):
def __init__(self, args, w2v):
super(TrainDataset, self).__init__()
self.train_data = torch.load(args.train_data)
self.w2id = {w: w2v.wv.vocab[w].index for w in w2v.wv.vocab}
self.pad_idx = w2v.wv.vocab.get("<pad>").index
self.maxlen = args.maxlen
self.maxlen_seed = args.maxlen_seed
self.w2v = w2v
self.maxlen_seed_aspects = args.maxlen_seed_aspects
self.general_idx = args.general_idx
self.num_cluster_general = args.num_cluster_general
self.num_cluster_aspects = args.num_cluster_aspects
self.all_seed, self.num_clusters, self.num_arr = self.load_file(args.seed_data)
def load_file(self, file: str):
seeds = []
num_clusters = []
len_dcts = []
dt = []
with open(file, "r") as f:
for line in f:
if len(line.strip().split()) > 0:
dt.append(line.strip())
# Take seed aspects-general
dt = dt = [
" ".join(dt[i].split()[: self.maxlen_seed])
if i == self.general_idx
else " ".join(dt[i].split()[: self.maxlen_seed_aspects])
for i in range(len(dt))
]
# --------END---------------
tmp = []
for item in dt:
tmp.extend(item.split())
union = []
l = Counter(tmp)
for wd, nm in l.items():
if nm > 1:
union.append(wd)
for i in range(len(dt)):
num_clus = -1
if i == self.general_idx:
num_clus = self.num_cluster_general
else:
num_clus = self.num_cluster_aspects
num_max = 0
for sd in dt[i].split():
if sd in self.w2v.wv.vocab:
num_max += 1
if num_clus > num_max:
num_clus = num_max
s = dt[i].split()
dct = dict([(i, []) for i in range(num_clus)])
X = np.asarray([self.w2v[w] for w in s if w in self.w2v.wv.vocab])
kmedoids = KMedoids(n_clusters=num_clus, random_state=0).fit(X)
for i in range(len(X)):
if (s[i] not in union) and (s[i] in self.w2v.wv.vocab):
dct[kmedoids.labels_[i]].append(s[i])
len_dct = 0
seed = []
for k, v in dct.items():
if len(v) > 0:
len_dct += 1
seed.append(" ".join(v))
seeds.extend(seed)
dcts = [len_dct] * len(seed)
len_dcts.extend(dcts)
num_clusters.append(len_dct)
return seeds, num_clusters, len_dcts
def sent2idx(self, tokens: List[str], is_sent):
# tokens = tokens.split()
if is_sent:
tokens = [
self.w2id[token.strip()]
for token in tokens
if token in self.w2id.keys()
]
tokens += [self.pad_idx] * (self.maxlen - len(tokens))
tokens = tokens[: self.maxlen]
else:
tokens = [
self.w2id[token.strip()]
for token in tokens
if token in self.w2id.keys()
]
ratio = int(self.maxlen_seed / len(tokens)) + 1
tokens = tokens * ratio
tokens = tokens[: self.maxlen_seed]
return tokens
def __len__(self):
return len(self.train_data)
def __getitem__(self, id):
sent = self.sent2idx(self.train_data[id][0], 1)
self.mask = [i != self.pad_idx for i in sent]
if np.array(self.mask).max() == 0:
self.mask = np.concatenate(([1], np.zeros(len(self.mask) - 1)))
seeds = []
for item in self.all_seed:
seeds.append(self.sent2idx(item.split(), 0))
return (
torch.LongTensor(sent),
torch.LongTensor(seeds),
torch.as_tensor(self.num_clusters),
torch.as_tensor(self.num_arr),
torch.as_tensor(np.array(self.mask).astype(int)),
) # , torch.LongTensor(label), torch.tensor(reli)
class TestDataset(data.Dataset):
def __init__(self, args, w2v):
super(TestDataset, self).__init__()
self.test_data = torch.load(args.test_data)
self.maxlen = args.maxlen
self.maxlen_seed = args.maxlen_seed
if "<pad>" in w2v.wv.vocab:
self.pad_idx = w2v.wv.vocab.get("<pad>").index
else:
self.pad_idx = w2v.wv.vocab.get("etc").index
self.w2id = {w: w2v.wv.vocab[w].index for w in w2v.wv.vocab}
self.general_idx = args.general_idx
self.num_cluster_general = args.num_cluster_general
self.num_cluster_aspects = args.num_cluster_aspects
self.w2v = w2v
self.maxlen_seed_aspects = args.maxlen_seed_aspects
self.all_seed, self.num_clusters, self.num_arr = self.load_file(args.seed_data)
def load_file(self, file: str):
seeds = []
num_clusters = []
len_dcts = []
dt = []
with open(file, "r") as f:
for line in f:
if len(line.strip().split()) > 0:
dt.append(line.strip())
# Take seed aspects-general
dt = dt = [
" ".join(dt[i].split()[: self.maxlen_seed])
if i == self.general_idx
else " ".join(dt[i].split()[: self.maxlen_seed_aspects])
for i in range(len(dt))
]
# --------END---------------
tmp = []
for item in dt:
tmp.extend(item.split())
union = []
l = Counter(tmp)
for wd, nm in l.items():
if nm > 1:
union.append(wd)
for i in range(len(dt)):
num_clus = -1
if i == self.general_idx:
num_clus = self.num_cluster_general
num_max = self.num_cluster_general
else:
num_clus = self.num_cluster_aspects
num_max = 0
for sd in dt[i].split():
if sd in self.w2v.wv.vocab:
num_max += 1
if num_clus > num_max:
num_clus = num_max
s = dt[i].split()
dct = dict([(i, []) for i in range(num_clus)])
X = np.asarray([self.w2v[w] for w in s if w in self.w2v.wv.vocab])
kmedoids = KMedoids(n_clusters=num_clus, random_state=0).fit(X)
for i in range(len(X)):
if (s[i] not in union) and (s[i] in self.w2v.wv.vocab):
dct[kmedoids.labels_[i]].append(s[i])
len_dct = 0
seed = []
# print(dct)
for k, v in dct.items():
if len(v) > 0:
len_dct += 1
seed.append(" ".join(v))
seeds.extend(seed)
dcts = [len_dct] * len(seed)
len_dcts.extend(dcts)
num_clusters.append(len_dct)
return seeds, num_clusters, len_dcts
def sent2idx(self, tokens: List[str], is_sent):
if is_sent:
tokens = [
self.w2id[token.strip()]
for token in tokens
if token in self.w2id.keys()
]
tokens += [self.pad_idx] * (self.maxlen - len(tokens))
tokens = tokens[: self.maxlen]
else:
tokens = [
self.w2id[token.strip()]
for token in tokens
if token in self.w2id.keys()
]
ratio = int(self.maxlen_seed / len(tokens)) + 1
tokens = tokens * ratio
tokens = tokens[: self.maxlen_seed]
return tokens
def __len__(self):
return len(self.test_data)
def __getitem__(self, id):
sent = self.sent2idx(self.test_data[id][0].split(), 1)
self.mask = [i != self.pad_idx for i in sent]
if np.array(self.mask).max() == 0:
self.mask = np.concatenate(([1], np.zeros(len(self.mask) - 1)))
seeds = []
for item in self.all_seed:
seeds.append(self.sent2idx(item.split(), 0))
idx_sd = self.sent2idx(item.split(), 0)
label = self.test_data[id][1]
return (
torch.LongTensor(sent),
torch.LongTensor(seeds),
torch.as_tensor(label),
torch.as_tensor(self.num_clusters),
torch.as_tensor(self.num_arr),
torch.as_tensor(np.array(self.mask).astype(int)),
)