forked from theeluwin/pytorch-sgns
-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataset.py
209 lines (185 loc) · 9.24 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
import os
import codecs
import pickle
import datetime
import numpy as np
import pandas as pd
import json
import csv
import logging
def random_split(lst, frac=0.2):
np.random.seed(0)
return np.random.choice(lst, int(frac * len(lst)))
def split_usr_itms(itms_lsts):
return [usr_itms[:-1] for usr_itms in itms_lsts], itms_lsts
def filter(data_dict, min_count, max_count):
return {k: data_dict[k] for k in data_dict.keys() if min_count < len(data_dict[k]) < max_count}
class Preprocess(object):
def __init__(self, unk='<UNK>', pad='pad', raw_data_file='./data/', save_data_dir='./corpus/', line_sep=',',
pos_thresh=4, user_pos=0, item_pos=1, rate_pos=2, date_pos=3, min_usr_len=2, max_usr_len=1000,
min_items_cnt=5, max_items_cnt=50000, final_usr_len=4, split_strategy='leave_one_out'):
self.unk = unk
self.pad = pad
self.wc = {}
self.idx2item = list()
self.item2idx = dict()
self.vocab = set()
self.line_sep = line_sep
self.pos_thresh = pos_thresh
self.user_pos = user_pos
self.item_pos = item_pos
self.rate_pos = rate_pos
self.date_pos = date_pos
self.min_usr_len = min_usr_len
self.max_usr_len = max_usr_len
self.min_items_cnt = min_items_cnt
self.max_items_cnt = max_items_cnt
self.final_usr_len = final_usr_len
self.split_strategy = split_strategy
self.raw_data_file = raw_data_file
self.save_data_dir = save_data_dir
def build(self, filepath, ic_out, vocab_out, idx2item_out, item2idx_out):
logging.info("building vocab...")
step = 0
with codecs.open(filepath, 'r', encoding='utf-8') as file:
for line in file:
step += 1
if not step % 1000:
logging.info("working on {}kth line".format(step // 1000), end='\r')
line = line.strip()
if not line:
continue
user = line.split()
for item in user:
self.wc[item] = self.wc.get(item, 0) + 1
# sorted list of items in a descent order of their frequency
self.wc[self.unk] = 1
self.wc[self.pad] = 1
self.idx2item = sorted(self.wc, key=self.wc.get, reverse=True)
self.item2idx = {self.idx2item[idx]: idx for idx, _ in enumerate(self.idx2item)}
self.vocab = set([item for item in self.item2idx])
pickle.dump(self.wc, open(os.path.join(self.save_data_dir, ic_out), 'wb'))
pickle.dump(self.vocab, open(os.path.join(self.save_data_dir, vocab_out), 'wb'))
pickle.dump(self.idx2item, open(os.path.join(self.save_data_dir, idx2item_out), 'wb'))
pickle.dump(self.item2idx, open(os.path.join(self.save_data_dir, item2idx_out), 'wb'))
logging.info("build done")
def create_train_samp(self, user, item_target):
sub_user = user[:item_target]
target_item = user[item_target]
return [self.item2idx[item] for item in sub_user], self.item2idx[target_item]
def convert(self, filepath, savepath, train=False):
logging.info("converting corpus...")
step = 0
data = []
usrs_len = []
with codecs.open(filepath, 'r', encoding='utf-8') as file:
num_users = 0
for line in file:
step += 1
if not step % 1000:
logging.info("working on {}kth line".format(step // 1000), end='\r')
line = line.strip()
if not line:
continue
user = []
for item in line.split():
if item in self.vocab:
user.append(item)
else:
user.append(self.unk)
usrs_len.append(len(user))
num_users += 1
for item_target in range(1, len(user)):
if not train and item_target < (len(user) - 1):
continue
data.append((self.create_train_samp(user, item_target)))
print("")
pickle.dump(data, open(savepath, 'wb'))
logging.info("conversion done")
logging.info("num of users:", num_users)
logging.info("max user:", max(usrs_len))
def read_data(self, raw_file):
user2data = {}
item2data = {}
with open(raw_file) as rating_file:
for i, line in enumerate(rating_file):
if i % 5000000 == 0:
print(i)
if line != "\n":
line = line.strip().split(self.line_sep)
line = [i for i in line if i != '']
user_id = line[self.user_pos]
if user_id not in user2data:
user2data[user_id] = []
item_id = line[self.item_pos]
if item_id not in item2data:
item2data[item_id] = []
try:
date = int(line[self.date_pos])
except:
date = int(datetime.datetime.strptime(line[self.date_pos], '%Y-%m-%d').timestamp())
if float(line[self.rate_pos]) > self.pos_thresh:
user2data[user_id].append((line[self.item_pos], date))
item2data[item_id].append(line[self.user_pos])
return user2data, item2data
def split(self, users):
if self.split_strategy == 'users_split':
train_users = random_split(list(users.keys()))
full_train, test = {user: users[user] for user in train_users}, \
{user: users[user] for user in list(users.keys()) if user not in train_users}
train_users = random_split(list(full_train.keys()))
train, valid = {user: full_train[user] for user in train_users}, \
{user: full_train[user] for user in list(users.keys()) if user not in train_users}
elif self.split_strategy == 'leave_one_out':
full_train, test = {user: users[user][:-1] for user in users}, users
train, valid = {user: full_train[user][:-1] for user in full_train}, full_train
else:
print('Split strategy not valid')
return
return [full_train[user] for user in list(full_train.keys())], \
[train[user] for user in list(train.keys())], \
[valid[user] for user in list(valid.keys())], \
[test[user] for user in list(test.keys())]
def save_file(self, file_name, data):
with open(os.path.join(self.save_data_dir, file_name), 'w', newline="") as x:
csv.writer(x, delimiter=" ").writerows(data)
def generate_train_files(data_cnfg):
with open(data_cnfg) as f:
params = json.load(f)
preprocess = Preprocess(**params)
user2data, item2data = preprocess.read_data(preprocess.raw_data_file)
# filter user and items
user2data = filter(user2data, preprocess.min_usr_len, preprocess.max_usr_len)
item2data = {item: [user for user in item2data[item] if user in user2data.keys()] for item in item2data.keys()}
item2data = filter(item2data, preprocess.min_items_cnt, preprocess.max_items_cnt)
user2data = {user: [item for item in user2data[user] if item[0] in item2data.keys()] for user in user2data.keys()}
user2data = filter(user2data, preprocess.final_usr_len, preprocess.max_usr_len)
# arrange users data by date
user2data = {usr: [item_index[0] for item_index in sorted(user2data[usr], key=lambda x: x[1])] for usr in user2data.keys()}
# generate processed raw files
full_corpus = [user2data[user] for user in user2data.keys()]
pd.DataFrame({'user': list(user2data.keys()), 'item': [user2data[usr][-1] for usr in user2data.keys()]}).to_csv(
os.path.join(preprocess.save_data_dir, 'test_raw.csv'), header=False, index=False)
full_train, train, valid, test = preprocess.split(user2data)
preprocess.save_file('full_corpus.txt', full_corpus)
preprocess.save_file('full_train.txt', full_train)
preprocess.save_file('train.txt', train)
preprocess.save_file('valid.txt', valid)
preprocess.save_file('test.txt', test)
# generate final train files
preprocess.build(os.path.join(preprocess.save_data_dir, 'full_corpus.txt'), 'full_ic.dat', 'full_vocab.dat',
'full_idx2item.dat', f'full_item2idx.dat')
preprocess.build(os.path.join(preprocess.save_data_dir, 'full_train.txt'), 'ic.dat', 'vocab.dat',
'idx2item.dat', 'item2idx.dat')
print("Full train")
preprocess.convert(os.path.join(preprocess.save_data_dir, 'full_train.txt'),
os.path.join(preprocess.save_data_dir, 'full_train.dat'), train=True)
print("Test")
preprocess.convert(os.path.join(preprocess.save_data_dir, 'test.txt'),
os.path.join(preprocess.save_data_dir, 'test.dat'))
print("Train")
preprocess.convert(os.path.join(preprocess.save_data_dir, 'train.txt'),
os.path.join(preprocess.save_data_dir, 'train.dat'), train=True)
print("valid")
preprocess.convert(os.path.join(preprocess.save_data_dir, 'valid.txt'),
os.path.join(preprocess.save_data_dir, 'valid.dat'))