-
Notifications
You must be signed in to change notification settings - Fork 16
/
DataLoader.py
275 lines (233 loc) · 9.45 KB
/
DataLoader.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
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
''' Data Loader class for training iteration '''
import random
import numpy as np
import torch
from torch.autograd import Variable
import Constants
import logging
import pickle
class Options(object):
def __init__(self, data_name = 'twitter'):
#data options.
#data_name = 'twitter'
#train file path.
self.train_data = 'data/'+data_name+'/cascade.txt'
#valid file path.
self.valid_data = 'data/'+data_name+'/cascadevalid.txt'
#test file path.
self.test_data = 'data/'+data_name+'/cascadetest.txt'
self.u2idx_dict = 'data/'+data_name+'/u2idx.pickle'
self.idx2u_dict = 'data/'+data_name+'/idx2u.pickle'
#save path.
self.save_path = ''
self.batch_size = 32
self.net_data = 'data/'+data_name+'/edges.txt'
self.embed_dim = 64
self.embed_file = 'data/'+data_name+'/dw'+str(self.embed_dim)+'.txt'
class DataLoader(object):
''' For data iteration '''
def __init__(
self, data_name, data=0, load_dict=True, cuda=True, batch_size=32, shuffle=True, test=False, with_EOS=True, loadNE=True): #data = 0 for train, 1 for valid, 2 for test
self.options = Options(data_name)
self.options.batch_size = batch_size
self._u2idx = {}
self._idx2u = []
self.data = data
self.test = test
self.with_EOS = with_EOS
if not load_dict:
self._buildIndex()
with open(self.options.u2idx_dict, 'wb') as handle:
pickle.dump(self._u2idx, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open(self.options.idx2u_dict, 'wb') as handle:
pickle.dump(self._idx2u, handle, protocol=pickle.HIGHEST_PROTOCOL)
else:
with open(self.options.u2idx_dict, 'rb') as handle:
self._u2idx = pickle.load(handle)
with open(self.options.idx2u_dict, 'rb') as handle:
self._idx2u = pickle.load(handle)
self.user_size = len(self._u2idx)
self._train_cascades,train_len = self._readFromFile(self.options.train_data)
self._valid_cascades,valid_len = self._readFromFile(self.options.valid_data)
self._test_cascades,test_len = self._readFromFile(self.options.test_data)
self.train_size = len(self._train_cascades)
self.valid_size = len(self._valid_cascades)
self.test_size = len(self._test_cascades)
print("training set size:%d valid set size:%d testing set size:%d" % (self.train_size, self.valid_size, self.test_size))
print(self.train_size+self.valid_size+self.test_size)
print((train_len+valid_len+test_len+0.0)/(self.train_size+self.valid_size+self.test_size))
print(self.user_size-2)
self.cuda = cuda
#self.test = test
if self.data == 0:
self._n_batch = int(np.ceil(len(self._train_cascades) / batch_size))
elif self.data == 1:
self._n_batch = int(np.ceil(len(self._valid_cascades) / batch_size))
else:
self._n_batch = int(np.ceil(len(self._test_cascades) / batch_size))
self._batch_size = self.options.batch_size
self._iter_count = 0
if loadNE:
self._adj_list = self._readNet(self.options.net_data)
self._adj_dict_list=self._readNet_dict_list(self.options.net_data)
self._embeds = self._load_ne(self.options.embed_file,self.options.embed_dim)
self._need_shuffle = shuffle
if self._need_shuffle:
random.shuffle(self._train_cascades)
def _buildIndex(self):
#compute an index of the users that appear at least once in the training and testing cascades.
opts = self.options
train_user_set = set()
valid_user_set = set()
test_user_set = set()
lineid=0
for line in open(opts.train_data):
lineid+=1
if len(line.strip()) == 0:
continue
chunks = line.strip().split()
for chunk in chunks:
try:
user, timestamp = chunk.split(',')
except:
print(line)
print(chunk)
print(lineid)
train_user_set.add(user)
for line in open(opts.valid_data):
if len(line.strip()) == 0:
continue
chunks = line.strip().split()
for chunk in chunks:
user, timestamp = chunk.split(',')
valid_user_set.add(user)
for line in open(opts.test_data):
if len(line.strip()) == 0:
continue
chunks = line.strip().split()
for chunk in chunks:
user, timestamp = chunk.split(',')
test_user_set.add(user)
user_set = train_user_set | valid_user_set | test_user_set
pos = 0
self._u2idx['<blank>'] = pos
self._idx2u.append('<blank>')
pos += 1
self._u2idx['</s>'] = pos
self._idx2u.append('</s>')
pos += 1
for user in user_set:
self._u2idx[user] = pos
self._idx2u.append(user)
pos += 1
opts.user_size = len(user_set) + 2
self.user_size = len(user_set) + 2
print("user_size : %d" % (opts.user_size))
def _readNet(self, filename):
adj_list=[[],[],[]]
n_edges = 0
# add self edges
for i in range(self.user_size):
adj_list[0].append(i)
adj_list[1].append(i)
adj_list[2].append(1)
for line in open(filename):
if len(line.strip()) == 0:
continue
nodes = line.strip().split(',')
if nodes[0] not in self._u2idx.keys() or nodes[1] not in self._u2idx.keys():
continue
n_edges+=1
adj_list[0].append(self._u2idx[nodes[0]])
adj_list[1].append(self._u2idx[nodes[1]])
adj_list[2].append(1) # weight
print('edge:')
print(n_edges/2)
return adj_list
def _readNet_dict_list(self, filename):
adj_list={}
# add self edges
for i in range(self.user_size):
adj_list.setdefault(i,[i]) # [i] or []
for line in open(filename):
if len(line.strip()) == 0:
continue
nodes = line.strip().split(',')
if nodes[0] not in self._u2idx.keys() or nodes[1] not in self._u2idx.keys():
continue
adj_list[self._u2idx[nodes[0]]].append(self._u2idx[nodes[1]])
adj_list[self._u2idx[nodes[1]]].append(self._u2idx[nodes[0]])
return adj_list
def _load_ne(self, filename, dim):
embed_file=open(filename,'r')
line = embed_file.readline().strip()
dim = int(line.split()[1])
embeds = np.zeros((self.user_size,dim))
for line in embed_file.readlines():
line=line.strip().split()
if line[0] not in self._u2idx.keys():
print("cannot find this user")
print(line[0])
continue
embeds[self._u2idx[line[0]],:]= np.array(line[1:])
return embeds
def _readFromFile(self, filename):
"""read all cascade from training or testing files. """
total_len = 0
t_cascades = []
for line in open(filename):
if len(line.strip()) == 0:
continue
userlist = []
chunks = line.strip().split()
for chunk in chunks:
try:
user, timestamp = chunk.split(',')
except:
print(chunk)
if user in self._u2idx:
userlist.append(self._u2idx[user])
if len(userlist) > 1 and len(userlist)<=500:
total_len+=len(userlist)
if self.with_EOS:
userlist.append(Constants.EOS)
t_cascades.append(userlist)
return t_cascades,total_len
def __iter__(self):
return self
def __next__(self):
return self.next()
def __len__(self):
return self._n_batch
def next(self):
''' Get the next batch '''
def pad_to_longest(insts):
''' Pad the instance to the max seq length in batch '''
max_len = max(len(inst) for inst in insts)
inst_data = np.array([
inst + [Constants.PAD] * (max_len - len(inst))
for inst in insts])
inst_data_tensor = Variable(
torch.LongTensor(inst_data), volatile=self.test)
if self.cuda:
inst_data_tensor = inst_data_tensor.cuda()
return inst_data_tensor
if self._iter_count < self._n_batch:
batch_idx = self._iter_count
self._iter_count += 1
start_idx = batch_idx * self._batch_size
end_idx = (batch_idx + 1) * self._batch_size
if self.data == 0:
seq_insts = self._train_cascades[start_idx:end_idx]
elif self.data == 1:
seq_insts = self._valid_cascades[start_idx:end_idx]
else:
seq_insts = self._test_cascades[start_idx:end_idx]
seq_data = pad_to_longest(seq_insts)
return seq_data
else:
if self._need_shuffle:
random.shuffle(self._train_cascades)
#random.shuffle(self._test_cascades)
self._iter_count = 0
raise StopIteration()