forked from tdinh15/rvsm-rgsm-admm
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
270 lines (220 loc) · 10.5 KB
/
main.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
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
# -*- coding: utf-8 -*-
"""
main pgd resnet
Usage: python main_pgd_resNet.py
"""
import argparse
import os
import shutil
import time
import numpy as np
import torch.backends.cudnn as cudnn
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
from torch.autograd import Variable
import torch.nn.functional as F
import torch
import torch.nn as nn
import math
from prune_utils import *
from loss_func import *
from utils import *
from models import *
parser = argparse.ArgumentParser(description='PyTorch Cifar10 Training')
parser.add_argument('--epochs', default=200, type=int, metavar='N', help='number of total epochs to run')
parser.add_argument('--start-epoch', default=0, type=int, metavar='N', help='manual epoch number (useful on restarts)')
parser.add_argument('-b', '--batch-size', default=128, type=int, metavar='N',
help='mini-batch size (default: 128),only used for train')
parser.add_argument('--lr', '--learning-rate', default=0.1, type=float, metavar='LR', help='initial learning rate')
parser.add_argument('--dev', '--device', default=0, type=int, metavar='N', help='GPU to run on')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M', help='momentum')
parser.add_argument('--weight-decay', '--wd', default=5e-4, type=float, metavar='W',
help='weight decay (default: 5e-4)')
# Parameters for RVSM (beta, lamb) and RGSM (beta, lamb1, lamb2)
parser.add_argument('--beta', default=1e-2, type=float, help='beta value')
parser.add_argument('--lamb', default=1e-6, type=float, help='lambda value')
parser.add_argument('--beta1', default=1, type=float, help='beta value')
parser.add_argument('--lamb1', default=1e-2, type=float, help='lambda1 value')
parser.add_argument('--lamb2', default=1e-5, type=float, help='lambda2 value')
# Parameters for ADMM
parser.add_argument('--pcen', default=80, type=float, help='pruning percentage')
parser.add_argument('--sparsity', default='elem', type=str, metavar='M', help='type of sparsity pruning')
parser.add_argument('--method', default='default', type=str, help='type of training: default, rvsm, rgsm, admm')
parser.add_argument('--print-freq', '-p', default=10, type=int, metavar='N', help='print frequency (default: 10)')
parser.add_argument('--resume', default='', type=str, metavar='PATH', help='path to latest checkpoint (default: none)')
parser.add_argument('-e', '--evaluate', dest='evaluate', action='store_true', help='evaluate model on validation set')
if __name__ == '__main__':
global best_acc
best_acc = 0.0
start_epoch = 0
args = parser.parse_args()
device = 'cuda:'+str(args.dev) if torch.cuda.is_available() else 'cpu'
try:
os.mkdir('weights')
except:
pass
#--------------------------------------------------------------------------
# Load Network type
#--------------------------------------------------------------------------
net = MobileNet()
net = net.to(device)
#--------------------------------------------------------------------------
# Load Cifar data
#--------------------------------------------------------------------------
print('==> Preparing data...')
root = './data'
download = True
#normalize = transforms.Normalize(mean=[0.507, 0.487, 0.441], std=[0.267, 0.256, 0.276])
train_set = torchvision.datasets.CIFAR10(root=root,
train=True,
download=download,
transform=transforms.Compose([
transforms.RandomCrop(32, padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
#normalize,
]))
test_set = torchvision.datasets.CIFAR10(root=root,
train=False,
download=download,
transform=transforms.Compose([
transforms.ToTensor(),
#normalize,
]))
kwargs = {'num_workers':4, 'pin_memory':True}
batchsize_test = int(len(test_set)/100)
print('Batch size of the test set: ', batchsize_test)
test_loader = torch.utils.data.DataLoader(dataset=test_set, batch_size=batchsize_test, shuffle=False, **kwargs)
batchsize_train = 128
print('Batch size of the train set: ', batchsize_train)
train_loader = torch.utils.data.DataLoader(dataset=train_set, batch_size=batchsize_train, shuffle=True, **kwargs)
criterion = nn.CrossEntropyLoss()
#--------------------------------------------------------------------------
# Declare Hyper-parameteres
#--------------------------------------------------------------------------
method = args.method
# RVSM
beta = args.beta
lamb = args.lamb
gamma = np.sqrt(2*lamb/beta)
# RGSM
beta1 = args.beta1
lamb1 = args.lamb1
lamb2 = args.lamb2
# ADMM
rho = 1e-2
pcen = args.pcen
sparsity = args.sparsity
#--------------------------------------------------------------------------
# Training Process
#--------------------------------------------------------------------------
nepoch = 200
for epoch in range(nepoch):
print('Epoch ID', epoch)
if epoch < 80:
lr = 0.1
elif epoch < 120:
lr = 0.1/10
elif epoch < 160:
lr = 0.1/10/10
else:
lr = 0.1/10/10/10
optimizer = optim.SGD(net.parameters(), lr=lr, momentum=0.9, weight_decay=5e-4, nesterov=True)
#----------------------------------------------------------------------
# Training
#----------------------------------------------------------------------
correct = 0; total = 0; train_loss = 0
net.train()
if method != 'default':
weights = [p for n,p in net.named_parameters() if 'weight' in n and len(p.size())==4]
if method == 'rvsm':
sp_wt = [thres(w,gamma) for w in weights]
if method == 'rgsm':
sp_wt = rgsm_l0(weights, lamb1, beta1)
if method == 'admm':
zs = [projection(w.data,pcen,sparsity) for w in weights]
us = [torch.zeros_like(z) for z in zs]
for batch_idx, (x, target) in enumerate(train_loader):
optimizer.zero_grad()
x, target = x.to(device), target.to(device)
score = net(x)
if method == 'rgsm':
loss = criterion(score, target) + gpls(weights, lamb2)
else:
loss = criterion(score, target)
if method == 'admm':
for j in range(len(weights)):
zs[j] = weights[j].data + us[j]
zs[j] = projection(zs[j],pcen,sparsity)
us[j] += weights[j] - zs[j]
loss.backward()
if method != 'default':
for j in range(len(weights)):
if method == 'rvsm':
weights[j].data -= lr*beta*(weights[j].data-sp_wt[j].data)
if method == 'rgsm':
weights[j].data -= lr*beta1*(weights[j].data-sp_wt[j].data)
if method == 'admm':
for j in range(len(weights)):
weights[j].data -= lr*rho*(weights[j].data-zs[j]+us[j])
optimizer.step()
if method == 'rvsm':
sp_wt = [thres(w,gamma) for w in weights]
if method == 'rgsm':
sp_wt = rgsm_l0(weights, lamb1, beta1)
train_loss += loss.item()
_, predicted = torch.max(score.data, 1)
total += target.size(0)
correct += predicted.eq(target.data).cpu().sum()
progress_bar(batch_idx, len(train_loader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (train_loss/(batch_idx+1), 100.*correct.numpy()/total, correct, total))
#----------------------------------------------------------------------
# Testing
#----------------------------------------------------------------------
correct = 0; total = 0; test_loss = 0
net.eval()
if method == 'rvsm':
for wt in weights:
replace_weight(wt,gamma)
if method == 'rgsm':
for j in range(len(sp_wt)):
k_sp = sp_wt[j]
k_wt = weights[j]
k_sp.data, k_wt.data = k_wt.data, k_sp.data
with torch.no_grad():
for batch_idx, (x, target) in enumerate(test_loader):
x, target = x.to(device), target.to(device)
try:
score, pert_x = net(x, target)
except:
score = net(x)
loss = criterion(score, target)
test_loss += loss.item()
_, predicted = torch.max(score.data, 1)
total += target.size(0)
correct += predicted.eq(target.data).cpu().sum()
progress_bar(batch_idx, len(test_loader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (test_loss/(batch_idx+1), 100.*correct.numpy()/total, correct, total))
#----------------------------------------------------------------------
# Save the checkpoint
#----------------------------------------------------------------------
if method == 'default':
out = ''
if method == 'rvsm':
out = 'beta'+str(beta)+'lamb'+str(lamb)
if method == 'rgsm':
out = 'beta'+str(beta1)+'lamb1_'+str(lamb1)+'lamb2_'+str(lamb2)
if method == 'admm':
out = 'pcen'+str(pcen)+str(sparsity)
acc = 100.*correct.numpy()/total
if acc > best_acc:
print('Saving model...')
state = {
'net': net,
'acc': acc,
'epoch': epoch,
}
torch.save(state, './weights/'+str(method)+out+'.pth')
best_acc = acc
print('The best acc: ', best_acc)