-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathpretrainDataset.py
180 lines (155 loc) · 7.48 KB
/
pretrainDataset.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
from collections import defaultdict
import random
from torch.utils.data import DataLoader, Dataset
import torch
import os
import pickle
from tqdm import tqdm
from termcolor import cprint
import numpy as np
class PretrainDataloader:
def __init__(self, config):
super(PretrainDataloader, self).__init__()
self.item_count = -1
self.user_count = -1
self.valid_time = None
self.test_time = None
self.train_list = None
self.valid_list = None
self.test_list = None
self.load_data(config)
def load_data(self, config):
local_path = config['local_save_path'] + config['dataset']
if os.path.isfile(local_path + '/train.pkl'):
cprint("Load data from Pickle", 'red')
self.train_list = pickle.load(open(local_path + '/train.pkl', 'rb'))
self.test_list = pickle.load(open(local_path + '/test.pkl', 'rb'))
self.valid_list = pickle.load(open(local_path + '/valid.pkl', 'rb'))
info_dic = pickle.load(open(local_path + '/info.pkl', 'rb'))
self.item_count = info_dic['item_count']
self.user_count = info_dic['user_count']
else:
cprint("Load and process from .txt")
train_list = self.read_txt(config['data_path'] + config['dataset'] + '/' + config['dataset'] + '.train.inter')
valid_list = self.read_txt(config['data_path'] + config['dataset'] + '/' + config['dataset'] + '.valid.inter')
test_list = self.read_txt(config['data_path'] + config['dataset'] + '/' + config['dataset'] + '.test.inter')
self.item_count = self.item_count + 1
self.user_count = self.user_count + 1
self.train_list = train_list
self.valid_list = valid_list
self.test_list = test_list
self.save_data_pkl(config['local_save_path'] + config['dataset'])
config['item_count'] = self.item_count
config['user_count'] = self.user_count
config['n_train_examples'] = len(self.train_list)
config['n_valid_examples'] = len(self.valid_list)
config['n_test_examples'] = len(self.test_list)
if config['use_text_emb']:
text_emb = pickle.load(open(config['data_path'] + config['dataset'] + '/' + config['dataset'] + '.feat1CLS', 'rb'))
text_emb = torch.cat([torch.zeros([1, 768]), text_emb], dim=0)
config['text_emb'] = text_emb
assert self.item_count == text_emb.size()[0]
def read_txt(self, file):
examples = []
with open(file, 'r') as rf:
rf.readline()
for line in tqdm(rf):
line = line.strip('\n')
if len(line) < 0:
break
line_list = line.split('\t')
# item id都需要+1， 为padding token留位置
user_id = int(line_list[0])
item_seq = line_list[1].split(' ')
item_seq = [int(item)+1 for item in item_seq]
timestamp_seq = line_list[2].split(' ')
timestamp_seq = [int(t) for t in timestamp_seq]
target_item = int(line_list[3]) + 1
target_timestamp = int(line_list[4])
examples.append([user_id, item_seq, timestamp_seq, target_item, target_timestamp])
self.item_count = max(self.item_count, max(item_seq), target_item)
self.user_count = max(self.user_count, user_id)
return examples
def generate_dataloader(self, config):
train_dataset = PretrainDataset(self.train_list, 'train', config)
valid_dataset = PretrainDataset(self.valid_list, 'valid', config)
test_dataset = PretrainDataset(self.test_list, 'test', config)
train_dataloader = DataLoader(train_dataset, batch_size=config['train_batch_size'], collate_fn=train_dataset.collate_fn, shuffle=True)
valid_dataloader = DataLoader(valid_dataset, batch_size=config['test_batch_size'], collate_fn=valid_dataset.collate_fn, shuffle=True)
test_dataloader = DataLoader(test_dataset, batch_size=config['test_batch_size'], collate_fn=test_dataset.collate_fn, shuffle=False)
return train_dataloader, valid_dataloader, test_dataloader
def save_data_pkl(self, root_path):
if not os.path.exists(root_path):
os.makedirs(root_path)
cprint("Save Pickled data", 'red')
pickle.dump(self.train_list, open(root_path + '/train.pkl', 'wb'))
pickle.dump(self.test_list, open(root_path + '/test.pkl', 'wb'))
pickle.dump(self.valid_list, open(root_path + '/valid.pkl', 'wb'))
pickle.dump({
"item_count": self.item_count,
"user_count": self.user_count,
}, open(root_path + '/info.pkl', 'wb'))
class PretrainDataset(Dataset):
def __init__(self, example_list, mode, config):
"""
"""
super(PretrainDataset, self).__init__()
self.example_list = example_list
self.mode = mode
self.item_count = config['item_count']
self.pad = 0
self.length = len(self.example_list)
self.seq_max_len = config['max_seq_len']
self.pad_mode = config['pad_mode']
self.neg_count = config['neg_count'] if mode == 'train' else config['test_neg_count']
self.use_text_emb = config['use_text_emb']
self.config = config
if self.mode == 'valid' and type(self.neg_count) is int and self.config['stage'] == 'pretrain':
self.neg_sample = torch.randint(low=0, high=self.item_count - 1, size=(len(self.example_list), self.neg_count)).tolist()
assert len(self.neg_sample) == len(self.example_list)
else:
self.neg_sample = None
def __len__(self):
return self.length
def __getitem__(self, index):
user_id = self.example_list[index][0]
item_seq = self.example_list[index][1][-self.seq_max_len:]
timestamp_seq = self.example_list[index][2][-self.seq_max_len:]
label = self.example_list[index][3]
target_timestamp = self.example_list[index][4]
if self.neg_sample is not None:
negs = self.neg_sample[index][: self.neg_count]
else:
negs = [0]
return [0] * (self.seq_max_len - len(item_seq)) + item_seq, \
user_id, \
label, \
negs, \
len(item_seq), \
[0] * (self.seq_max_len - len(timestamp_seq)) + timestamp_seq, \
target_timestamp
def collate_fn(self, batch):
item_seqs = []
users_id = []
labels = []
negs = []
lengths = []
timestamp_seqs = []
target_timestamp = []
for example in batch:
item_seqs.append(example[0])
users_id.append(example[1])
labels.append(example[2])
negs.append(example[3])
lengths.append(example[4])
timestamp_seqs.append(example[5])
target_timestamp.append(example[6])
return {
'item_seqs': torch.tensor(item_seqs),
'users': torch.tensor(users_id),
'labels': torch.tensor(labels),
'neg_items': torch.tensor(negs),
'lengths': torch.tensor(lengths),
'timestamps': torch.tensor(timestamp_seqs),
'target_timestamp': torch.tensor(target_timestamp),
}