-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataset.py
133 lines (119 loc) · 5.69 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
import numpy as np
from tqdm import tqdm
class Dataset:
def __init__(self, num_items, user_seq, timestamps=None, idx_user_map=None, idx_item_map=None):
super().__init__()
# item 0 is for padding
self.num_items = num_items
self.user_seq = user_seq
self.num_users = len(self.user_seq)
if timestamps:
self.timestamps = timestamps
if idx_item_map:
self.idx_item_map = idx_item_map
if idx_user_map:
self.idx_user_map = idx_user_map
def __len__(self):
return len(self.user_seq)
def __getitem__(self, idx):
if hasattr(self, "timestamps"):
return self.user_seq[idx], self.timestamps[idx]
else:
return self.user_seq[idx]
class TrainingDataset(Dataset):
def __init__(self, num_items, user_seq, timestamps=None, idx_user_map=None, idx_item_map=None, max_seq_len=None):
super().__init__(num_items, user_seq, timestamps, idx_user_map, idx_item_map)
if max_seq_len:
self.max_seq_len = max_seq_len
self.split()
def split(self):
self.train_seq = []
self.train_targets = []
if hasattr(self, "timestamps"):
self.train_seq_timestamps = []
print("building training dataset...")
for idx in tqdm(range(self.num_users)):
seq = self.user_seq[idx]
if len(seq) < 4:
continue
if hasattr(self, "timestamps"):
seq_timestamps = self.timestamps[idx]
last_pos = len(seq) - 1
if hasattr(self, "max_seq_len"):
for b in range((last_pos + self.max_seq_len - 1) // self.max_seq_len):
if (last_pos - b * self.max_seq_len) > self.max_seq_len * 1.1:
self.train_targets.append(
seq[(last_pos - (b + 1) * self.max_seq_len):(last_pos - b * self.max_seq_len)])
self.train_seq.append(
seq[(last_pos - (b + 1) * self.max_seq_len - 1):(last_pos - b * self.max_seq_len - 1)])
if hasattr(self, "timestamps"):
self.train_seq_timestamps.append(seq_timestamps[(
last_pos - (
b + 1) * self.max_seq_len - 1):(
last_pos - b * self.max_seq_len - 1)])
else:
self.train_targets.append(
seq[1:(last_pos - b * self.max_seq_len)])
self.train_seq.append(
seq[0:(last_pos - b * self.max_seq_len - 1)])
if hasattr(self, "timestamps"):
self.train_seq_timestamps.append(
seq_timestamps[0:(last_pos - b * self.max_seq_len - 1)])
break
else:
self.train_targets.append(seq[1:last_pos])
self.train_seq.append(seq[0:last_pos - 1])
if hasattr(self, "timestamps"):
self.train_seq_timestamps.append(
seq_timestamps[0:last_pos - 1])
def __len__(self):
return len(self.train_seq)
def __getitem__(self, idx):
if hasattr(self, "timestamps"):
return self.train_seq[idx], self.train_targets[idx], self.train_seq_timestamps[idx]
else:
return self.train_seq[idx], self.train_targets[idx]
class EvaluationDataset(Dataset):
def __init__(self, num_items, user_seq, timestamps=None, idx_user_map=None, idx_item_map=None, max_seq_len=None,
num_negatives=100):
super().__init__(num_items, user_seq, timestamps, idx_user_map, idx_item_map)
if max_seq_len:
self.max_seq_len = max_seq_len
self.num_negatives = num_negatives
self.split()
def split(self):
self.eval_seq = []
self.eval_targets = []
self.eval_neg_samples = []
if hasattr(self, "timestamps"):
self.eval_seq_timestamps = []
print("building evaluation dataset...")
for idx in tqdm(range(self.num_users)):
seq = self.user_seq[idx]
if len(seq) < 4:
continue
last_pos = len(seq) - 1
self.eval_targets.append(seq[last_pos:last_pos + 1])
neg_candidate_set = list(set(range(1, self.num_items)) - set(seq))
neg_targets = np.random.choice(
neg_candidate_set, size=self.num_negatives)
self.eval_neg_samples.append(neg_targets)
if hasattr(self, "timestamps"):
seq_timestamps = self.timestamps[idx]
if hasattr(self, "max_seq_len"):
self.eval_seq.append(
seq[max(0, last_pos - self.max_seq_len):last_pos])
if hasattr(self, "timestamps"):
self.eval_seq_timestamps.append(
seq_timestamps[max(0, last_pos - self.max_seq_len):last_pos])
else:
self.eval_seq.append(seq[0:last_pos])
if hasattr(self, "timestamps"):
self.eval_seq_timestamps.append(seq_timestamps[0:last_pos])
def __len__(self):
return len(self.eval_seq)
def __getitem__(self, idx):
if hasattr(self, "timestamps"):
return self.eval_seq[idx], self.eval_targets[idx], self.eval_seq_timestamps[idx], self.eval_neg_samples[idx]
else:
return self.eval_seq[idx], self.eval_targets[idx], self.eval_neg_samples[idx]