-
Notifications
You must be signed in to change notification settings - Fork 2
/
preprocess.py
472 lines (363 loc) · 15 KB
/
preprocess.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
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
import json
import argparse
import json
from collections import defaultdict
from time import mktime, strptime
import networkx as nx
import pandas as pd
import scipy.sparse as sp
from tqdm import tqdm
from utils import *
def load_and_save_yelp():
# keys = ['user_id', 'name', 'review_count', 'yelping_since', 'useful',
# 'funny', 'cool', 'elite', 'friends', 'fans', 'average_stars',
# 'compliment_hot', 'compliment_more', 'compliment_profile', 'compliment_cute',
# 'compliment_list', 'compliment_note', 'compliment_plain', 'compliment_cool',
# 'compliment_funny', 'compliment_writer', 'compliment_photos']
# Yelp download url: https://www.kaggle.com/yelp-dataset/yelp-dataset
user_path = r'datasets/Yelp/yelp_academic_dataset_user.json'
u2u_list = list()
with open(user_path, 'rb') as f:
for line in tqdm(f.readlines()):
line = json.loads(line)
uid = line['user_id']
friends = line['friends'].split(', ')
for fid in friends:
u2u_list.append([uid, fid])
rating_path = r'datasets/Yelp/yelp_academic_dataset_review.json'
u2i_list = list()
with open(rating_path, 'rb') as f:
for line in tqdm(f.readlines()):
line = json.loads(line)
uid = line['user_id']
iid = line['business_id']
rate = line['stars']
ts = int(mktime(strptime(line['date'], "%Y-%m-%d %H:%M:%S")))
u2i_list.append([uid, iid, ts, rate])
print('u2u =', len(u2u_list))
print('u2i =', len(u2i_list))
uid_map = defaultdict(int) # str id --> int id
iid_map = defaultdict(int)
uid_map[0] = 0
iid_map[0] = 0
user_num = 1
item_num = 1
for i, (uid, iid, ts, rate) in tqdm(enumerate(u2i_list)):
if uid_map[uid] == 0:
uid_map[uid] = user_num
user_num += 1
if iid_map[iid] == 0:
iid_map[iid] = item_num
item_num += 1
u2i_list[i] = [uid_map[uid], iid_map[iid], ts, rate]
u2i = np.array(u2i_list, dtype=np.int)
u2i = u2i[np.argsort(u2i[:, 0])] # sort by user id
new_u2u_list = list()
for u1, u2 in u2u_list:
new_u1, new_u2 = uid_map[u1], uid_map[u2]
if new_u1 and new_u2:
new_u2u_list.append([new_u1, new_u2])
u2u = np.array(new_u2u_list, dtype=np.int)
u2u = u2u[np.argsort(u2u[:, 0])] # sort by u1 id
print('min uid =', np.min(u2i[:, 0]))
print('max uid =', np.max(u2i[:, 0]))
print('num uid =', len(np.unique(u2i[:, 0])))
print('min iid =', np.min(u2i[:, 1]))
print('max iid =', np.max(u2i[:, 1]))
print('num iid =', len(np.unique(u2i[:, 1])))
print('min ts =', np.min(u2i[:, 2]))
print('max ts =', np.max(u2i[:, 2]))
print('min rate =', np.min(u2i[:, 3]))
print('max rate =', np.max(u2i[:, 3]))
print('num rate =', len(np.unique(u2i[:, 3])))
print('min u1 id =', np.min(u2u[:, 0]))
print('max u1 id =', np.max(u2u[:, 0]))
print('num u1 id =', len(np.unique(u2u[:, 0])))
print('min u2 id =', np.min(u2u[:, 1]))
print('max u2 id =', np.max(u2u[:, 1]))
print('num u2 id =', len(np.unique(u2u[:, 1])))
print(u2i[:50])
print(u2u[:50])
np.savez(file='datasets/Yelp/u2ui.npz',
u2i=u2i,
u2u=u2u)
np.savez(file='datasets/Yelp/uid_map.npz',
uid_map=uid_map)
print('saved at', 'datasets/Yelp/u2ui.npz')
def filter_and_reid():
'''
Raw u2i (8021100, 3)
min user = 1
max user = 1968703
num user = 1968703
min item = 1
max item = 209393
num item = 209393
Raw u2u edges: 19042100
'''
u2ui = np.load(f'datasets/Yelp/u2ui.npz')
u2u, u2i = u2ui['u2u'], u2ui['u2i']
df = pd.DataFrame(data=u2i, columns=['user', 'item', 'ts', 'rate'])
df.drop_duplicates(subset=['user', 'item', 'ts', 'rate'], keep='first', inplace=True)
print('Raw u2i', df.shape)
print('min user =', df['user'].min())
print('max user =', df['user'].max())
print('num user =', len(np.unique(df.values[:, 0])))
print('min item =', df['item'].min())
print('max item =', df['item'].max())
print('num item =', len(np.unique(df.values[:, 1])))
df = preprocess_uir(df, prepro='3filter', pos_threshold=3, level='u')
df.drop(['rate'], axis=1, inplace=True)
print('Processed u2i', df.shape)
print('min user =', df['user'].min())
print('max user =', df['user'].max())
print('num user =', len(np.unique(df.values[:, 0])))
print('min item =', df['item'].min())
print('max item =', df['item'].max())
print('num item =', len(np.unique(df.values[:, 1])))
df = df.sort_values(['user', 'ts'], kind='mergesort').reset_index(drop=True)
u2i = df.values
user_idmap = defaultdict(int) # src id -> new id
user_idmap[0] = 0
user_num = 1
for i, (user, item, ts) in tqdm(enumerate(u2i)):
if user_idmap[user] == 0:
user_idmap[user] = user_num
user_num += 1
u2i[i, 0] = user_idmap[user]
print('Raw u2u edges:', len(u2u))
new_uu_elist = []
for u1, u2 in tqdm(u2u):
new_u1 = user_idmap[u1]
new_u2 = user_idmap[u2]
if new_u1 and new_u2:
new_uu_elist.append([new_u1, new_u2])
print('Processed u2u edges:', len(new_uu_elist))
u2u = np.array(new_uu_elist).astype(np.int32)
u2i = u2i.astype(np.int32)
save_path = 'datasets/Yelp/reid_u2ui.npz'
np.savez(file=save_path, u2u=u2u, u2i=u2i)
print('saved at', save_path)
def delete_isolated_user():
u2ub = np.load('datasets/Yelp/reid_u2ui.npz')
uu_elist = u2ub['u2u']
u2i = u2ub['u2i']
print('Building u2u graph...')
user_num = np.max(u2i[:, 0]) + 1
g = nx.Graph()
g.add_nodes_from(list(range(user_num)))
g.add_edges_from(uu_elist)
g.remove_node(0)
isolated_user_set = set(nx.isolates(g))
print('Isolated user =', len(isolated_user_set))
new_u2i = []
for user, item, ts in tqdm(u2i):
if user not in isolated_user_set:
new_u2i.append([user, item, ts])
new_u2i = np.array(new_u2i, dtype=np.int32)
print('No isolated user u2i =', new_u2i.shape)
user_idmap = defaultdict(int) # src id -> new id
user_idmap[0] = 0
user_num = 1
for i, (user, item, ts) in tqdm(enumerate(new_u2i)):
if user_idmap[user] == 0:
user_idmap[user] = user_num
user_num += 1
new_u2i[i, 0] = user_idmap[user]
new_uu_elist = []
for u1, u2 in tqdm(uu_elist):
new_u1 = user_idmap[u1]
new_u2 = user_idmap[u2]
if new_u1 and new_u2:
new_uu_elist.append([new_u1, new_u2])
new_uu_elist = np.array(new_uu_elist, dtype=np.int32)
df = pd.DataFrame(data=new_u2i, columns=['user', 'item', 'ts'])
df['item'] = pd.Categorical(df['item']).codes + 1
print(df.head(20))
# cc_sizes = [len(c) for c in sorted(nx.connected_components(g), key=len, reverse=True)]
# print('u2u connected components sizes (top20):', cc_sizes[:20])
# print('Isolated user =', np.sum(np.array(cc_sizes) == 1))
user_num = df['user'].max() + 1
item_num = df['item'].max() + 1
print('min user =', df['user'].min())
print('max user =', df['user'].max())
num_user = len(np.unique(df.values[:, 0]))
print('num user =', num_user)
print('min item =', df['item'].min())
print('max item =', df['item'].max())
num_item = len(np.unique(df.values[:, 1]))
print('num item =', num_item)
print(f'Loaded Yelp dataset with {user_num} users, {item_num} items, '
f'{len(df.values)} u2i, {len(new_uu_elist)} u2u. ')
new_u2i = df.values.astype(np.int32)
save_path = 'datasets/Yelp/noiso_reid_u2ui.npz'
np.savez(file=save_path, u2u=new_uu_elist, u2i=new_u2i)
return num_user, num_item
def data_partition(df):
print('Splitting train/val/test set...')
user_train = defaultdict(list)
user_valid = defaultdict(list)
user_test = defaultdict(list)
eval_users = []
valid_items = []
test_items = []
user_items_dict = defaultdict(list)
def apply_fn1(grp):
key_id = grp['user'].values[0]
user_items_dict[key_id] = grp[['item', 'ts']].values
df.groupby('user').apply(apply_fn1)
print('Groupby user finished.')
for user in tqdm(user_items_dict.keys()):
nfeedback = len(user_items_dict[user])
if nfeedback < 5:
user_train[user] = user_items_dict[user]
else:
# Append user history items
eval_users.append(user)
user_train[user] = user_items_dict[user][:-2]
# Second last item for validation
valid_item = user_items_dict[user][-2][0]
user_valid[user].append(valid_item)
valid_items.append(valid_item)
# Last item for test
test_item = user_items_dict[user][-1][0]
user_test[user].append(test_item)
test_items.append(test_item)
return user_train, user_valid, user_test, eval_users, valid_items, test_items
def gen_and_save_u2u_dict_and_split(num_user, num_item):
u2ui = np.load('datasets/Yelp/noiso_reid_u2ui.npz')
print('Building u2u graph...')
g = nx.Graph()
g.add_nodes_from(list(range(num_user)))
g.add_edges_from(u2ui['u2u'])
g.remove_node(0)
print('To undirected graph...')
g.to_undirected()
# g.add_edges_from([[u, u] for u in g.nodes])
u2u_dict = nx.to_dict_of_lists(g)
df = pd.DataFrame(data=u2ui['u2i'], columns=['user', 'item', 'ts'])
print('Raw u2i =', df.shape)
df.drop_duplicates(subset=['user', 'item', 'ts'], keep='first', inplace=True)
df = df.sort_values(['user', 'ts'], kind='mergesort').reset_index(drop=True)
print('Processed u2i =', df.shape)
user_train, user_valid, user_test, eval_users, valid_items, test_items = data_partition(df)
save_path = 'datasets/Yelp/u2u_split_dicts.pkl'
save_pkl(save_path, [
u2u_dict, user_train, user_valid, user_test,
eval_users, valid_items, test_items])
print('saved at', save_path)
def get_nbr(u2u, user, nbr_maxlen):
nbr = np.zeros([nbr_maxlen, ], dtype=np.int32)
nbr_len = len(u2u[user])
if nbr_len == 0:
pass
elif nbr_len > nbr_maxlen:
np.random.shuffle(u2u[user])
nbr[:] = u2u[user][:nbr_maxlen]
else:
nbr[:nbr_len] = u2u[user]
return nbr
def get_nbr_iids(user_train, user, nbrs, time_splits):
nbr_maxlen = len(nbrs)
seq_maxlen = len(time_splits)
nbrs_iids = np.zeros((nbr_maxlen, seq_maxlen), dtype=np.int32)
start_idx = np.nonzero(time_splits)[0]
if len(start_idx) == 0:
return nbrs_iids
else:
start_idx = start_idx[0]
user_first_ts = time_splits[start_idx]
user_last_ts = time_splits[-1]
for i, nbr in enumerate(nbrs):
if nbr == 0 or nbr == user:
continue
nbr_hist = user_train[nbr]
if len(nbr_hist) == 0:
continue
nbr_first_ts = nbr_hist[0][1]
nbr_last_ts = nbr_hist[-1][1]
if nbr_first_ts > user_last_ts or nbr_last_ts <= user_first_ts:
continue
sample_list = list()
for j in range(start_idx + 1, seq_maxlen):
start_time = time_splits[j - 1]
end_time = time_splits[j]
if start_time != end_time:
sample_list = list(filter(None, map(
lambda x: x[0] if x[1] > start_time and x[1] <= end_time else None, nbr_hist
)))
if len(sample_list):
# print('st={} et={} sl={}'.format(start_time, end_time, sample_list))
nbrs_iids[i, j] = np.random.choice(sample_list)
return nbrs_iids
def gen_and_save_all_user_batches(user_num, item_num):
# eval batch for each user
u2u_dict, user_train, user_valid, user_test, eval_users, valid_items, test_items = \
load_pkl('datasets/Yelp/u2u_split_dicts.pkl')
def sample_one_user(user):
seq = np.zeros(seq_maxlen, dtype=np.int32)
pos = np.zeros(seq_maxlen, dtype=np.int32)
ts = np.zeros(seq_maxlen, dtype=np.int32)
nxt = user_train[user][-1, 0]
idx = seq_maxlen - 1
for (item, time_stamp) in reversed(user_train[user][:-1]):
seq[idx] = item
ts[idx] = time_stamp
pos[idx] = nxt
nxt = item
idx -= 1
if idx == -1: break
nbr = get_nbr(u2u_dict, user, nbr_maxlen)
nbr_iid = get_nbr_iids(user_train, user, nbr, ts)
nbr_iid = sp.csr_matrix(nbr_iid, dtype=np.int32)
return user, seq, pos, nbr, nbr_iid
uid_list = []
seq_list = []
pos_list = []
nbr_list = []
nbr_iid_list = []
for user in tqdm(range(1, user_num)):
user, seq, pos, nbr, nbr_iid = sample_one_user(user)
uid_list.append(user)
seq_list.append(seq)
pos_list.append(pos)
nbr_list.append(nbr)
nbr_iid_list.append(nbr_iid)
# save as npz
np.savez(
'datasets/Yelp/processed_data.npz',
user_train=user_train,
user_valid=user_valid,
user_test=user_test,
eval_users=np.array(eval_users, dtype=np.int32),
valid_items=np.array(valid_items, dtype=np.int32),
test_items=np.array(test_items, dtype=np.int32),
train_uid=np.array(uid_list, dtype=np.int32),
train_seq=np.array(seq_list, dtype=np.int32),
train_pos=np.array(pos_list, dtype=np.int32),
train_nbr=np.array(nbr_list, dtype=np.int32),
train_nbr_iid=nbr_iid_list
)
print('saved at datasets/Yelp/processed_data.npz')
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='')
parser.add_argument('--dataset', default='Yelp')
parser.add_argument('--edim', type=int, default=64)
parser.add_argument('--seq_maxlen', type=int, default=50)
parser.add_argument('--nbr_maxlen', type=int, default=20)
parser.add_argument('--batch_size', type=int, default=1024)
parser.add_argument('--seed', type=int, default=0)
args = parser.parse_args()
seed = args.seed
np.random.seed(seed)
seq_maxlen = args.seq_maxlen
nbr_maxlen = args.nbr_maxlen
load_and_save_yelp() # 从logs生成npz
# 1. 筛选数据,重新标id
# filter_and_reid() # 5filter删除item和user -> 重新赋值uid -> 从u2u删除没有item的user
# num_user, num_item = delete_isolated_user() # 删除孤立用户 -> 重新赋值uid -> 更新u2u和u2b的uid
# num_user = num_user + 1
# num_item = num_item + 1
# 2. Generate Training Samples
# gen_and_save_u2u_dict_and_split(num_user, num_item)
# gen_and_save_all_user_batches(num_user, num_item)