-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataloader.py
192 lines (157 loc) · 7.43 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
import numpy as np
import time
class DataSet(object):
def __init__(self, filename, args, mark_u_time_end, mark_v_time_end):
self.filename = filename
self.f = open(self.filename, 'r')
self.max_batch_size = args.max_batch_size
self.time_interval = 604800 * args.time_interval # day 86400, week 604800, month 30/31 days 2592000/2678400
self.finish = 0
self.num_of_u = args.num_of_u
self.num_of_v = args.num_of_v
self.gran_u = 604800 * args.gran_u
self.gran_v = 604800 * args.gran_v
self.mark_u_time_end = mark_u_time_end
self.mark_v_time_end = mark_v_time_end
def get_batch_data(self):
b = time.time()
all_data = []
inputs_u_idx = []
inputs_v_idx = []
inputs_idx_pair = []
last_pos = self.f.tell()
line = self.f.readline()
self.f.seek(last_pos)
data = line.rstrip('\n').split('\t')
start_time = float(data[3])
end_time = start_time + self.time_interval
for i in range(self.max_batch_size):
last_pos = self.f.tell()
line = self.f.readline()
if line != '':
data = line.rstrip('\n').split('\t')
if float(data[3]) <= end_time:
all_data.append(data)
inputs_u_idx.append(int(data[0]))
inputs_v_idx.append(int(data[1]))
inputs_idx_pair.append([int(data[0]), int(data[1])])
else:
self.f.seek(last_pos)
break
else:
self.finish = 1
self.f.close()
self.f = open(self.filename, 'r')
break
inputs_u_idx = np.sort(list(set(inputs_u_idx)))-1
inputs_v_idx = np.sort(list(set(inputs_v_idx)))-1
all_data = np.asarray(all_data, dtype=np.float32)
# Get mapped inputs_idx_pair
inputs_idx_pair = np.asarray(inputs_idx_pair)
siz = inputs_idx_pair.shape
tmp = np.concatenate((inputs_idx_pair, np.array(range(siz[0]), ndmin=2).T
, np.zeros([siz[0], 2]), np.expand_dims(all_data[:, 2], 1)), axis=1)
tmp = tmp[tmp[:, 0].argsort()]
# Map user indices
u_idx = -1
mark = 0
for i in range(siz[0]):
if mark == int(tmp[i, 0]):
tmp[i, 0] = u_idx
else:
mark = int(tmp[i, 0])
u_idx += 1
tmp[i, 0] = u_idx
tmp = tmp[tmp[:, 1].argsort()] # Reorder to the original order
# Map item indices
v_idx = -1
mark = 0
for i in range(siz[0]):
if mark == int(tmp[i, 1]):
tmp[i, 1] = v_idx
else:
mark = int(tmp[i, 1])
v_idx += 1
tmp[i, 1] = v_idx
tmp = tmp[tmp[:, 2].argsort()] # Reorder to the original order
u_time = np.zeros(inputs_u_idx.shape[0])-1
v_time = np.zeros(inputs_v_idx.shape[0])-1
mark_u_time = np.zeros(inputs_u_idx.shape[0])-self.gran_u-1
mark_v_time = np.zeros(inputs_v_idx.shape[0])-self.gran_v-1
mark_u_time_end = self.mark_u_time_end[inputs_u_idx]
mark_v_time_end = self.mark_v_time_end[inputs_v_idx]
sp_u_indices = []
sp_v_indices = []
sp_u_val = []
sp_v_val = []
sp_u_indices_res = []
sp_v_indices_res = []
sp_u_val_res = []
sp_v_val_res = []
for i in range(siz[0]): # siz[0]
if mark_u_time[int(tmp[i, 0])] + self.gran_u <= all_data[i, 3]:
u_time[int(tmp[i, 0])] += 1
mark_u_time[int(tmp[i, 0])] = all_data[i, 3]
tmp[i, 3] = u_time[int(tmp[i, 0])]
sp_u_indices.append([int(tmp[i, 0]), int(tmp[i, 3]), int(all_data[i, 1])-1])
sp_u_val.append(all_data[i, 2])
sp_u_indices_res.append([int(tmp[i, 0]), int(tmp[i, 3]), 0])
sp_u_val_res.append(all_data[i, 4])
if int(tmp[i, 3]) > 0:
sp_u_indices_res.append([int(tmp[i, 0]), int(tmp[i, 3])-1, 1])
sp_u_val_res.append(mark_u_time_end[int(tmp[i, 0])])
mark_u_time_end[int(tmp[i, 0])] = all_data[i, 3]
if all_data[i, 4] == 0:
sp_u_indices_res.append([int(tmp[i, 0]), int(tmp[i, 3]), 2])
sp_u_val_res.append(1)
else:
tmp[i, 3] = u_time[int(tmp[i, 0])]
sp_u_indices.append([int(tmp[i, 0]), int(tmp[i, 3]), int(all_data[i, 1])-1])
sp_u_val.append(all_data[i, 2])
mark_u_time_end[int(tmp[i, 0])] = all_data[i, 3]
if mark_v_time[int(tmp[i, 1])] + self.gran_v <= all_data[i, 3]:
v_time[int(tmp[i, 1])] += 1
mark_v_time[int(tmp[i, 1])] = all_data[i, 3]
tmp[i, 4] = v_time[int(tmp[i, 1])]
sp_v_indices.append([int(tmp[i, 1]), int(tmp[i, 4]), int(all_data[i, 0])-1])
sp_v_val.append(all_data[i, 2])
sp_v_indices_res.append([int(tmp[i, 1]), int(tmp[i, 4]), 0])
sp_v_val_res.append(all_data[i, 5])
if int(tmp[i, 4]) > 0:
sp_v_indices_res.append([int(tmp[i, 1]), int(tmp[i, 4])-1, 1])
sp_v_val_res.append(mark_v_time_end[int(tmp[i, 1])])
mark_v_time_end[int(tmp[i, 1])] = all_data[i, 3]
if all_data[i, 5] == 0:
sp_v_indices_res.append([int(tmp[i, 1]), int(tmp[i, 4]), 2])
sp_v_val_res.append(1)
else:
tmp[i, 4] = v_time[int(tmp[i, 1])]
sp_v_indices.append([int(tmp[i, 1]), int(tmp[i, 4]), int(all_data[i, 0])-1])
sp_v_val.append(all_data[i, 2])
mark_v_time_end[int(tmp[i, 1])] = all_data[i, 3]
u_mode = max(tmp[:, 3])+1
v_mode = max(tmp[:, 4])+1
for i in range(inputs_u_idx.shape[0]):
sp_u_indices_res.append([i, u_time[i], 1])
sp_u_val_res.append(mark_u_time_end[i])
for i in range(inputs_v_idx.shape[0]):
sp_v_indices_res.append([i, v_time[i], 1])
sp_v_val_res.append(mark_v_time_end[i])
self.mark_u_time_end[inputs_u_idx] = mark_u_time_end
self.mark_v_time_end[inputs_v_idx] = mark_v_time_end
sp_u_shape = np.array([inputs_u_idx.shape[0], u_mode, self.num_of_v])
sp_v_shape = np.array([inputs_v_idx.shape[0], v_mode, self.num_of_u])
sp_u_shape_res = np.array([inputs_u_idx.shape[0], u_mode, 3])
sp_v_shape_res = np.array([inputs_v_idx.shape[0], v_mode, 3])
sp_u_indices = np.asarray(sp_u_indices, dtype=np.int32)
sp_u_val = np.asarray(sp_u_val)
sp_u_indices_res = np.asarray(sp_u_indices_res, dtype=np.int32)
sp_u_val_res = np.asarray(sp_u_val_res)
sp_v_indices = np.asarray(sp_v_indices, dtype=np.int32)
sp_v_val = np.asarray(sp_v_val)
sp_v_indices_res = np.asarray(sp_v_indices_res, dtype=np.int32)
sp_v_val_res = np.asarray(sp_v_val_res)
print('Total Batch Cut Time:', time.time() - b)
return sp_u_indices, sp_u_shape, sp_u_val, sp_u_indices_res, sp_u_shape_res, sp_u_val_res,\
sp_v_indices, sp_v_shape, sp_v_val, sp_v_indices_res, sp_v_shape_res, sp_v_val_res,\
inputs_u_idx, inputs_v_idx, tmp[:, [0, 1, 3, 4, 5]], all_data