-
Notifications
You must be signed in to change notification settings - Fork 0
/
torch_utils.py
173 lines (129 loc) · 5.94 KB
/
torch_utils.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
import torch
import sys
import copy
import torchvision
import torch.nn as nn
import torchvision.transforms as transforms
from torchvision.datasets import SVHN, USPS, MNIST
from torch.utils.data import DataLoader
import numpy as np
import random
def config_gpus_setup(model, gpus='all', print_stat=False):
"""
:param model: pytorch model
:param gpus: list of gpus ids to use or 'all' to use all of them (default)
:param print_stat: if True - print the number of visible GPUs
:return: model, device
"""
ngpus = torch.cuda.device_count()
if print_stat:
print('Cuda see %s GPUs' % ngpus)
if gpus == 'all':
gpus = list(range(ngpus))
if torch.cuda.is_available():
device = torch.device("cuda")
model = model.to(device)
if (ngpus > 1):
model = torch.nn.DataParallel(model, device_ids=gpus)
else:
device = torch.device("cpu")
return model, device
def set_random_seeds(seed=123):
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
torch.backends.cudnn.enabled = False
torch.backends.cudnn.deterministic = True
def get_model(nclasses=10):
model = torchvision.models.resnet18()
# Finetune Final few layers to adjust for tiny imagenet input
model.avgpool = nn.AdaptiveAvgPool2d(1)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, nclasses)
return model
class DuplicateToChannels:
"""Duplicate single channel 3 times"""
def __init__(self):
pass
def __call__(self, x):
return x.repeat((3,1,1))
def get_svhn(split='train', resize=(224, 224), batch_size=64):
transform = transforms.Compose([transforms.Resize(resize), transforms.ToTensor()])
svhn = SVHN('../datasets/svhn', split=split, transform=transform)
svhn_loader = DataLoader(svhn, batch_size=batch_size, shuffle=True, num_workers=20)
return svhn_loader
def get_usps(split='train', resize=(224, 224), batch_size=64):
transform = transforms.Compose([transforms.Resize(resize), transforms.ToTensor(), DuplicateToChannels()])
usps = USPS('../datasets/usps', train=(split=='train'), transform=transform)
usps_loader = DataLoader(usps, batch_size=batch_size, shuffle=True, num_workers=20)
return usps_loader
def get_mnist(split='train', resize=(224, 224), batch_size=64):
transform = transforms.Compose([transforms.Resize(resize), transforms.ToTensor(), DuplicateToChannels()])
mnist = MNIST('../datasets/mnist', train=(split=='train'), transform=transform)
mnist_loader = DataLoader(mnist, batch_size=batch_size, shuffle=True, num_workers=20)
return mnist_loader
def get_device(model):
try:
device = model.device
except:
device = 'cuda'
return device
def train_model(model, train_loader, criterion, optimizer,
val_loader=None, scheduler=None, epochs=2, gpus='all',dataset_sizes=None, label_one_hot=False):
def batch_step(model, X, y, optimizer, scheduler):
optimizer.zero_grad()
y_pred = model.forward(X)
loss = criterion(y_pred, y)
if model.training:
loss.backward(retain_graph=True)
optimizer.step()
if scheduler is not None:
scheduler.step()
return y_pred, loss, optimizer, scheduler
def batch_metric_updates(X, y, y_pred, loss, running_loss, running_corrects, label_one_hot):
_, y_scalar = torch.max(y_pred, 1)
running_loss += loss.item() * X.size(0)
if label_one_hot:
_, y = torch.max(y, 1)
running_corrects += torch.sum(y_scalar == y.data)
return running_loss, running_corrects
def epoch_step(model, loader, optimizer, scheduler, metrics, label_one_hot):
running_loss = 0.0
running_corrects = 0
for i, (X, y) in enumerate(loader):
device = get_device(model)
X, y = X.to(device), y.to(device)
y_pred, loss, optimizer, scheduler = batch_step(model, X, y, optimizer, scheduler)
running_loss, running_corrects = batch_metric_updates(X, y, y_pred, loss, running_loss, running_corrects,
label_one_hot)
epoch_loss = running_loss / loader.dataset.data.shape[0]
epoch_acc = running_corrects.double() / loader.dataset.data.shape[0]
prefix = '' if model.training else 'val_'
metrics['%sloss' % prefix].append(epoch_loss)
metrics['%sacc' % prefix].append(epoch_acc)
return model, optimizer, scheduler, metrics
metrics = {'loss': [], 'acc': [], 'val_loss': [], 'val_acc': []}
best_model = copy.deepcopy(model.state_dict())
best_val_score = 0.0
for epoch in range(epochs):
model.train()
model, optimizer, scheduler, metrics = epoch_step(model, train_loader, optimizer, scheduler, metrics,
label_one_hot)
report = 'Epoch: %d, loss:%0.4f ,acc:%0.4f'%(epoch, metrics['loss'][-1],metrics['acc'][-1])
if val_loader is not None:
model.eval()
model, optimizer, scheduler, metrics = epoch_step(model, val_loader, optimizer, scheduler, metrics,
label_one_hot)
report = '%s ,val_loss:%0.4f ,val_acc:%0.4f'%(report, metrics['val_loss'][-1],metrics['val_acc'][-1])
if metrics['val_acc'][-1] > best_val_score:
best_val_score = metrics['val_acc'][-1]
best_model = copy.deepcopy(model.state_dict())
# model.load_state_dict(best_model_wts)
print(report)
return model, metrics, best_model
def print_and_log(txt, log_path):
print(txt)
f = open(log_path, 'a')
f.write(txt+'\n')
f.close()