-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_eval.py
300 lines (247 loc) · 10.5 KB
/
main_eval.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
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
#Parts of codes are copied and modified from "https://github.com/kuangliu/pytorch-cifar"
from __future__ import print_function
import numpy as np
import torch
import utils
import util
import torch.nn as nn
import torch.optim as optim
import torchvision.datasets as dset
import torch.backends.cudnn as cudnn
import torch.nn.functional as F
import torchvision.transforms as transforms
from util import progress_bar
from torch.utils.data import DataLoader
from torch.autograd import Variable
import pdb
import os
import argparse
import time
import struct
import random
parser = argparse.ArgumentParser(description='PyTorch CIFAR10 Training')
parser.add_argument('--lr', default=0.1, type=float, help='learning rate')
parser.add_argument('--resume', '-r', action='store_true', help='resume from checkpoint')
parser.add_argument('--se', default=0, type=int, help='start epoch')
parser.add_argument('--ne', default=0, type=int, help='number of epoch')
parser.add_argument('--pr', default=0, type=int, help='pruning') # mode=1 is pruning, mode=0 is no pruning
parser.add_argument('--bs', default=128, type=int, help='batch size')
parser.add_argument('--mode', default=0, type=int, help='train or inference') #mode=1 is train, mode=0 is inference
parser.add_argument('--network', default='ckpt_20200630_equalized.t0', help='input network ckpt name', metavar="FILE")
parser.add_argument('--output', default='garbage.txt', help='output file name', metavar="FILE")
parser.add_argument('--fixed', default=0, type=float)
parser.add_argument('--pprec', type=int, default=7, metavar='N',help='parameter precision for layer weight')
parser.add_argument('--aprec', type=int, default=7, metavar='N',help='arithmetic precision for convolution and fc')
parser.add_argument('--iwidth', type=int, default=6, metavar='N',help='integer bitwidth for internal part')
args = parser.parse_args()
torch.set_printoptions(precision=16)
use_cuda = torch.cuda.is_available()
best_acc = 0 # best test accuracy
transform_train = transforms.Compose([transforms.RandomCrop(32,padding=4),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225])])
transform_test = transforms.Compose([transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225])])
cifar_train = dset.CIFAR100("/home/yhbyun/Dataset/CIFAR100/", train=True, transform=transform_train, target_transform=None, download=True)
cifar_test = dset.CIFAR100("/home/yhbyun/Dataset/CIFAR100/", train=False, transform=transform_test, target_transform=None, download=True)
train_loader = torch.utils.data.DataLoader(cifar_train,batch_size=args.bs, shuffle=True,num_workers=8,drop_last=False)
test_loader = torch.utils.data.DataLoader(cifar_test,batch_size=192, shuffle=False,num_workers=6,drop_last=False)
mode = args.mode
class QConv2d(nn.Conv2d):
def forward(self, input):
input = super().forward(input)
if args.fixed == 1:
return quant(input, args.iwidth, args.aprec)
else:
return input
class QBatchNorm2d(nn.BatchNorm2d):
def forward(self, input):
input = super().forward(input)
if args.fixed == 1:
return quant(input, args.iwidth, args.aprec)
else:
return input
class QLinear(nn.Linear):
def forward(self, input):
input = super().forward(input)
if args.fixed == 1:
return quant(input, args.iwidth, args.aprec)
else:
return input
class Block(nn.Module):
'''expand + depthwise + pointwise'''
def __init__(self, in_planes, out_planes, expansion, stride):
super(Block, self).__init__()
self.stride = stride
planes = expansion * in_planes
self.conv1 = QConv2d(in_planes, planes, kernel_size=1, stride=1, padding=0, bias=True)
self.bn1 = QBatchNorm2d(planes)
self.conv2 = QConv2d(planes, planes, kernel_size=3, stride=stride, padding=1, groups=planes, bias=True)
self.bn2 = QBatchNorm2d(planes)
self.conv3 = QConv2d(planes, out_planes, kernel_size=1, stride=1, padding=0, bias=True)
self.bn3 = QBatchNorm2d(out_planes)
self.shortcut = nn.Sequential()
if stride == 1 and in_planes != out_planes:
self.shortcut = nn.Sequential(
QConv2d(in_planes, out_planes, kernel_size=1, stride=1, padding=0, bias=True),
QBatchNorm2d(out_planes),
)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = F.relu(self.bn2(self.conv2(out)))
out = self.bn3(self.conv3(out))
out = out + self.shortcut(x) if self.stride==1 else out
return out
class MobileNetV2(nn.Module):
# (expansion, out_planes, num_blocks, stride)
cfg = [(1, 16, 1, 1),
(6, 24, 2, 1), # NOTE: change stride 2 -> 1 for CIFAR10
(6, 32, 3, 2),
(6, 64, 4, 2),
(6, 96, 3, 1),
(6, 160, 3, 2),
(6, 320, 1, 1)]
def __init__(self, num_classes=100):
super(MobileNetV2, self).__init__()
# NOTE: change conv1 stride 2 -> 1 for CIFAR10
self.conv1 = QConv2d(3, 32, kernel_size=3, stride=1, padding=1, bias=True)
self.bn1 = QBatchNorm2d(32)
self.layers = self._make_layers(in_planes=32)
self.conv2 = QConv2d(320, 1280, kernel_size=1, stride=1, padding=0, bias=True)
self.bn2 = QBatchNorm2d(1280)
self.linear = QLinear(1280, num_classes)
#self.fileoutindex = 0
def _make_layers(self, in_planes):
layers = []
for expansion, out_planes, num_blocks, stride in self.cfg:
strides = [stride] + [1]*(num_blocks-1)
for stride in strides:
layers.append(Block(in_planes, out_planes, expansion, stride))
in_planes = out_planes
return nn.Sequential(*layers)
def forward(self, x):
out = self.conv1(x)
out = F.relu(self.bn1(out))
out = self.layers(out)
out = F.relu(self.bn2(self.conv2(out)))
# NOTE: change pooling kernel_size 7 -> 4 for CIFAR10
out = F.avg_pool2d(out, 4)
out = out.view(out.size(0), -1)
out = self.linear(out)
return out
def quant(input, intbit, fracbit):
input = torch.div(torch.round(torch.mul(input,2 ** fracbit)),2 ** fracbit)
input = torch.clamp(input,-(2 ** intbit), 2 ** intbit - 2 ** (-fracbit))
return input
# Training
def train(net, epoch):
print('\nEpoch: %d' % epoch)
net.train()
train_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(train_loader):
if use_cuda:
inputs, targets = inputs.cuda(), targets.cuda()
optimizer.zero_grad()
inputs, targets = Variable(inputs), Variable(targets)
outputs = net(inputs)
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
train_loss += loss.data.item()
_, predicted = torch.max(outputs.data, 1)
total += float(targets.size(0))
correct += predicted.eq(targets.data).cpu().sum().type(torch.FloatTensor)
progress_bar(batch_idx, len(train_loader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (train_loss/(batch_idx+1), 100.*correct/total, correct, total))
def test(net):
global best_acc
net.eval()
test_loss = 0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(test_loader):
if use_cuda:
inputs, targets = inputs.cuda(), targets.cuda()
inputs, targets = Variable(inputs), Variable(targets)
outputs = net(inputs)
loss = criterion(outputs, targets)
test_loss += loss.data.item()
_, predicted = torch.max(outputs.data, 1)
total += float(targets.size(0))
correct += predicted.eq(targets.data).cpu().sum().type(torch.FloatTensor)
progress_bar(batch_idx, len(test_loader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (test_loss/(batch_idx+1), 100.*correct/total, correct, total))
# Save checkpoint.
acc = 100.*correct/total
return acc
def retrain(net, epoch):
print('\nEpoch: %d' % epoch)
global best_acc
net.train()
train_loss = 0
total = 0
correct = 0
for batch_idx, (inputs, targets) in enumerate(train_loader):
if use_cuda:
inputs, targets = inputs.cuda(), targets.cuda()
optimizer.zero_grad()
inputs, targets = Variable(inputs), Variable(targets)
outputs = net(inputs)
loss = criterion(outputs, targets)
loss.backward()
if args.fixed:
net = util.quantize(net, args.pprec)
optimizer.step()
train_loss += loss.data.item()
_, predicted = torch.max(outputs.data, 1)
total += targets.size(0)
correct += float(predicted.eq(targets.data).cpu().sum())
progress_bar(batch_idx, len(train_loader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
% (train_loss/(batch_idx+1), 100.*correct/total, correct, total))
acc = 100.*correct/total
if args.fixed:
net = util.quantize(net, args.pprec)
print('==> Resuming from checkpoint..')
checkpoint = torch.load('./checkpoint/'+args.network)
net = checkpoint['net']
checkpoint = torch.load('./checkpoint/ckpt_20190913.t0')
net_origin = checkpoint['net']
assert os.path.isdir('checkpoint'), 'Error: no checkpoint directory found!'
net2 = MobileNetV2()
mask_null = util.maskGen(net, isbias=1, isempty = 1)
#mask_null = util.maskGen(net, isbias=0, isempty = 1)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=args.lr, momentum=0.9, weight_decay=5e-4)
start_epoch = args.se
num_epoch = args.ne
if use_cuda:
net.cuda()
net = torch.nn.DataParallel(net, device_ids=range(torch.cuda.device_count()))
net2.cuda()
net2 = torch.nn.DataParallel(net2, device_ids=range(torch.cuda.device_count()))
net_origin.cuda()
net_origin = torch.nn.DataParallel(net_origin, device_ids=range(torch.cuda.device_count()))
cudnn.benchmark = True
params = util.paramsGet(net)
tmp = (params.data != 0).sum()
print("Ratio of nonzero value : ",tmp.item()/params.size()[0])
print("Number of nonzero value : ",tmp.item())
print("Number of value", params.size()[0])
net2 = util.netMaskMul(net2, mask_null, isbias = 1)
net2 = util.addNetwork(net2, net, isbias = 1)
net2 = util.swapBatch(net2, net)
if args.fixed:
net = util.quantize(net, args.pprec)
net2 = util.quantize(net2, args.pprec)
net_origin = util.quantize(net_origin, args.pprec)
def calcDiff(tensor):
pass
def evalMetric(net, net_origin):
params_net = utils.paramsGet(net)
params_net_origin = utils.paramsGet(net_origin)
print(params_net.size())
print(params_net_origin.size())
evalMetric()