-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathdataloader.py
147 lines (121 loc) · 6.12 KB
/
dataloader.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
from torchvision import datasets, transforms
import torch
import numpy as np
from imagenet32_dataset import ImageNet32
custom_data_path = {
'conic': 'data/conic.npz',
'mlp6': 'data/mlp.npz',
'mlp3': 'data/mlp3.npz',
'mlp9': 'data/mlp9.npz',
'CIFAR10': 'data/CIFAR10',
'CIFAR100': 'data/CIFAR100',
'MNIST': 'data/MNIST',
'ImageNet32': 'data/ImageNet32'
}
def get_dataset(dataset_name, train_samples, test_samples, batch_size, random_labels=False):
im_size, padded_im_size, num_classes, in_channels = dataset_stats(dataset_name)
target_trans = (lambda y: torch.randint(0, num_classes, (1,)).item()) if random_labels else None
if dataset_name == 'MNIST':
transform = transforms.Compose([transforms.Pad((padded_im_size - im_size)//2),
transforms.ToTensor(),
transforms.Normalize(0.1307,0.3081)])
train_set = datasets.MNIST(custom_data_path['MNIST'], train=True, download=True, transform=transform, target_transform=target_trans)
if train_samples:
train_set = torch.utils.data.Subset(train_set, range(train_samples))
train_loader = torch.utils.data.DataLoader(train_set, batch_size=batch_size, shuffle=True, num_workers=2)
test_set = datasets.MNIST(custom_data_path['MNIST'], train=False, download=True, transform=transform, target_transform=target_trans)
if test_samples:
test_set = torch.utils.data.Subset(test_set, range(test_samples))
test_loader = torch.utils.data.DataLoader(test_set, batch_size=batch_size, shuffle=True, num_workers=2)
in_channels = 1
num_classes = 10
if dataset_name == 'CIFAR10':
transform = transforms.Compose(
[transforms.Pad((padded_im_size - im_size)//2),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
train_set = datasets.CIFAR10(root=custom_data_path['CIFAR10'], train=True, download=True, transform=transform, target_transform=target_trans)
if train_samples:
train_set = torch.utils.data.Subset(train_set, range(train_samples))
train_loader = torch.utils.data.DataLoader(train_set, batch_size=batch_size, shuffle=True, num_workers=2)
test_set = datasets.CIFAR10(root=custom_data_path['CIFAR10'], train=False, download=True, transform=transform, target_transform=target_trans)
if test_samples:
test_set = torch.utils.data.Subset(test_set, range(test_samples))
test_loader = torch.utils.data.DataLoader(test_set, batch_size=batch_size, shuffle=True, num_workers=2)
in_channels = 3
num_classes = 10
if dataset_name == 'CIFAR100':
transform = transforms.Compose(
[transforms.Pad((padded_im_size - im_size)//2),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
train_set = datasets.CIFAR100(root=custom_data_path['CIFAR100'], train=True, download=True, transform=transform, target_transform=target_trans)
if train_samples:
train_set = torch.utils.data.Subset(train_set, range(train_samples))
train_loader = torch.utils.data.DataLoader(train_set, batch_size=batch_size, shuffle=True, num_workers=8)
test_set = datasets.CIFAR100(root=custom_data_path['CIFAR100'], train=False, download=True, transform=transform, target_transform=target_trans)
if test_samples:
test_set = torch.utils.data.Subset(test_set, range(test_samples))
test_loader = torch.utils.data.DataLoader(test_set, batch_size=batch_size, shuffle=True, num_workers=8)
in_channels = 3
num_classes = 100
if dataset_name == 'ImageNet32':
transform = transforms.Compose(
[transforms.Pad((padded_im_size - im_size)//2),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))])
train_set = ImageNet32(root=custom_data_path['ImageNet32'], num_samples=train_samples, train=True, transform=transform, target_transform=target_trans)
test_set = ImageNet32(root=custom_data_path['ImageNet32'], num_samples=test_samples, train=False, transform=transform, target_transform=target_trans)
train_loader = torch.utils.data.DataLoader(train_set, batch_size=batch_size)
test_loader = torch.utils.data.DataLoader(test_set, batch_size=batch_size)
in_channels = 3
num_classes = 1000
if dataset_name == 'conic' or dataset_name[:3] == 'mlp':
print(f'Using custom dataset at {custom_data_path[dataset_name]}')
npz = np.load(custom_data_path[dataset_name])
X, y = npz['X'], npz['y']
if random_labels:
y = torch.randint(0, num_classes, y.shape)
tensor_X, tensor_y = torch.tensor(X).float(), torch.tensor(y).to(torch.int64)
train_loader = torch.utils.data.DataLoader(
torch.utils.data.TensorDataset(tensor_X[:train_samples, :], tensor_y[:train_samples]),
batch_size=batch_size, shuffle=True)
test_loader = torch.utils.data.DataLoader(
torch.utils.data.TensorDataset(tensor_X[train_samples: train_samples + test_samples, :], tensor_y[train_samples: train_samples + test_samples]),
batch_size=batch_size, shuffle=True)
in_channels = None
return train_loader, test_loader, num_classes, in_channels
def dataset_stats(dataset):
if dataset == 'MNIST':
# dataset parameters
im_size = 28
padded_im_size = 32
num_classes = 10
input_ch = 1
elif dataset == 'CIFAR10':
# dataset parameters
im_size = 32
padded_im_size = 32
num_classes = 10
input_ch = 3
elif dataset == 'CIFAR100':
# dataset parameters
im_size = 32
padded_im_size = 32
num_classes = 100
input_ch = 3
elif dataset == 'ImageNet32':
# dataset parameters
im_size = 32
padded_im_size = 32
num_classes = 1000
input_ch = 3
elif dataset == 'conic' or dataset.startswith('mlp'):
# dataset parameters
im_size = None
padded_im_size = None
num_classes = 4
input_ch = None
else:
raise ValueError('Dataset not recognized.')
return im_size, padded_im_size, num_classes, input_ch