-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsaver_loader.py
149 lines (101 loc) · 4.16 KB
/
saver_loader.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
import pickle
import numpy as np
import random
class DataSaver:
def __init__(self, save_path, comment=None):
self.file = open(save_path, "wb")
self.comment = comment
self.path_list = []
self.discription_list = []
self.block_num = 0
self.file.write(np.uint64(self.block_num).tobytes())
self.current_pos = self.file.tell()
def save(self, element_list):
if not all([type(element) is np.ndarray for element in element_list]):
raise Exception("Wrong data type")
self.path_list.append(self.current_pos)
discription = []
for element in element_list:
bytes = element.tobytes()
self.file.write(bytes)
size = len(bytes)
discription.append((element.shape, element.dtype, size))
self.discription_list.append(discription)
self.current_pos = self.file.tell()
bytes_annotation = pickle.dumps((self.path_list, self.discription_list, self.comment))
self.file.write(bytes_annotation)
self.file.seek(0, 0)
self.file.write(np.uint64(self.current_pos).tobytes())
self.file.seek(self.current_pos, 0)
def close(self):
self.file.close()
class DataLoader:
def __init__(self, path, shuffle=False, load_into_memory=False, max_items=None, random_seed=False):
self.file = open(path, "rb")
self.file.seek(0, 2)
self.file_size = self.file.tell()
self.file.seek(0, 0)
uint64_size = 8
annotation_pos = np.fromstring(self.file.read(uint64_size), dtype=np.uint64)[0]
self.file.seek(annotation_pos, 0)
bytes_annotation = self.file.read()
self.path_list, self.discription_list, self.comment = pickle.loads(bytes_annotation)
if max_items != None:
self.max_items = max_items if max_items < len(self.path_list) else len(self.path_list)
self.path_list = self.path_list[:self.max_items]
self.discription_list = self.path_list[:self.max_items]
else:
self.max_items = len(self.path_list)
self.load_into_memory = load_into_memory
if load_into_memory:
self.shuffle = False
self.index = 0
self.file.seek(self.path_list[self.index])
self.data_in_memory = []
for idx in range(self.max_items):
self.data_in_memory.append(self.load_from_file())
self.shuffle = shuffle
if shuffle:
self.shuffle_data()
self.index = random.randint(0, self.max_items - 1) if random_seed else 0
self.file.seek(self.path_list[self.index])
def getItem(self):
if self.load_into_memory:
self.reload_if_needed()
item = self.data_in_memory[self.index]
self.index += 1
return item
else:
return self.load_from_file()
def load_from_file(self):
self.reload_if_needed()
if self.shuffle:
self.file.seek(self.path_list[self.index], 0)
discriptions = self.discription_list[self.index]
total_size = sum([size for _, _, size in discriptions])
bytes_data = self.file.read(total_size)
item = [None] * len(discriptions)
current_pos = 0
for idx, (shape, dtype, size) in enumerate(discriptions):
bytes_element = bytes_data[current_pos : current_pos + size]
element = np.fromstring(bytes_element, dtype=dtype)
item[idx] = np.reshape(element, shape)
current_pos += size
self.index += 1
return item
def reload_if_needed(self):
if self.index >= self.max_items:
self.index = 0
if self.shuffle:
self.shuffle_data()
def __len__(self):
return self.max_items
def getComment(self):
return self.comment
def shuffle_data(self):
if self.load_into_memory:
random.shuffle(self.data_in_memory)
else:
combined = list(zip(self.path_list, self.discription_list))
random.shuffle(combined)
self.path_list[:], self.discription_list[:] = zip(*combined)