-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_util.py
251 lines (209 loc) · 9.43 KB
/
train_util.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
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
import time
import torch
import torchvision.transforms as transforms
import torchvision
import torch.nn as nn
import torch.optim as optim
from prune import prune
from pruned_layers import *
device = 'cuda' if torch.cuda.is_available() else 'cpu'
def train(net, epochs=100, batch_size=128, lr=0.01, reg=5e-4, prn=False):
"""
Training a network
:param net:
:param epochs:
:param batch_size:
"""
print('==> Preparing data..')
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
best_acc = 0 # best test accuracy
start_epoch = 0 # start from epoch 0 or last checkpoint epoch
trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform_train)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, num_workers=16)
testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform_test)
testloader = torch.utils.data.DataLoader(testset, batch_size=100, shuffle=False, num_workers=2)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=lr, momentum=0.9, weight_decay=reg, nesterov=False)
scheduler = optim.lr_scheduler.MultiStepLR(optimizer, milestones=[int(epochs*0.5), int(epochs*0.75)], gamma=0.1)
global_steps = 0
start = time.time()
for epoch in range(start_epoch, epochs):
"""
Start the training code.
"""
print('\nEpoch: %d' % epoch)
net.train()
train_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(trainloader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
outputs = net(inputs)
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
if prn == True:
prune(net, method='percentage', q=85.0, s=0.75)
train_loss += loss.item()
_, predicted = outputs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
global_steps += 1
if global_steps % 16 == 0:
end = time.time()
num_examples_per_second = 16 * batch_size / (end - start)
print("[Step=%d]\tLoss=%.4f\tacc=%.4f\t%.1f examples/second"
% (global_steps, train_loss / (batch_idx + 1), (correct / total), num_examples_per_second))
start = time.time()
scheduler.step()
"""
Start the testing code.
"""
net.eval()
test_loss = 0
correct = 0
total = 0
with torch.no_grad():
for batch_idx, (inputs, targets) in enumerate(testloader):
inputs, targets = inputs.to(device), targets.to(device)
outputs = net(inputs)
loss = criterion(outputs, targets)
test_loss += loss.item()
_, predicted = outputs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
num_val_steps = len(testloader)
val_acc = correct / total
print("Test Loss=%.4f, Test acc=%.4f" % (test_loss / (num_val_steps), val_acc))
if val_acc > best_acc:
best_acc = val_acc
if prn == False:
print("Saving to net_before_pruning")
torch.save(net.state_dict(), "net_before_pruning.pt")
if prn == True:
print("Saving to net_after_pruning")
torch.save(net.state_dict(), "net_after_pruning.pt")
def finetune_after_prune(net, epochs=100, batch_size=128, lr=0.01, reg=5e-4):
print('==> Preparing data..')
transform_train = transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
best_acc = 0 # best test accuracy
start_epoch = 0 # start from epoch 0 or last checkpoint epoch
trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform_train)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, num_workers=16)
testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform_test)
testloader = torch.utils.data.DataLoader(testset, batch_size=100, shuffle=False, num_workers=2)
net = net.to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=lr, momentum=0.9, weight_decay=reg, nesterov=False)
global_steps = 0
start = time.time()
for epoch in range(start_epoch, epochs):
"""
Start the training code.
"""
print('\nEpoch: %d' % epoch)
net.train()
train_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(trainloader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
outputs = net(inputs)
loss = criterion(outputs, targets)
loss.backward()
# before optimizer.step(), manipulate the gradient
"""
Zero the gradients of the pruned variables.
-----------------------Your Code-------------------------
"""
assert isinstance(net, nn.Module)
for n, m in net.named_modules():
if isinstance(m, PrunedConv):
first_grads = m.conv.weight.grad.cpu().numpy()
new_grads = np.multiply(first_grads, m.mask)
m.conv.weight.grad = torch.from_numpy(new_grads).float().to(device)
elif isinstance(m, PruneLinear):
first_grads = m.linear.weight.grad.cpu().numpy()
new_grads = np.multiply(first_grads, m.mask)
m.linear.weight.grad = torch.from_numpy(new_grads).float().to(device)
optimizer.step()
train_loss += loss.item()
_, predicted = outputs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
global_steps += 1
if global_steps % 16 == 0:
end = time.time()
num_examples_per_second = 16 * batch_size / (end - start)
print("[Step=%d]\tLoss=%.4f\tacc=%.4f\t%.1f examples/second"
% (global_steps, train_loss / (batch_idx + 1), (correct / total), num_examples_per_second))
start = time.time()
"""
Start the testing code.
"""
net.eval()
test_loss = 0
correct = 0
total = 0
with torch.no_grad():
for batch_idx, (inputs, targets) in enumerate(testloader):
inputs, targets = inputs.to(device), targets.to(device)
outputs = net(inputs)
loss = criterion(outputs, targets)
test_loss += loss.item()
_, predicted = outputs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
num_val_steps = len(testloader)
val_acc = correct / total
print("Test Loss=%.4f, Test acc=%.4f" % (test_loss / (num_val_steps), val_acc))
if val_acc > best_acc:
best_acc = val_acc
print("Saving...")
torch.save(net.state_dict(), "net_after_pruning1.pt")
def test(net):
transform_test = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010)),
])
testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform_test)
testloader = torch.utils.data.DataLoader(testset, batch_size=100, shuffle=False, num_workers=2)
classes = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
criterion = nn.CrossEntropyLoss()
net.eval()
test_loss = 0
correct = 0
total = 0
with torch.no_grad():
for batch_idx, (inputs, targets) in enumerate(testloader):
inputs, targets = inputs.to(device), targets.to(device)
outputs = net(inputs)
loss = criterion(outputs, targets)
test_loss += loss.item()
_, predicted = outputs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
num_val_steps = len(testloader)
val_acc = correct / total
print("Test Loss=%.4f, Test accuracy=%.4f" % (test_loss / (num_val_steps), val_acc))
return val_acc