-
Notifications
You must be signed in to change notification settings - Fork 2
/
vis_pyhessian_analysis.py
233 lines (191 loc) · 8.42 KB
/
vis_pyhessian_analysis.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
#*
# @file Different utility functions
# Copyright (c) Zhewei Yao, Amir Gholami
# All rights reserved.
# This file is part of PyHessian library.
#
# PyHessian is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# PyHessian is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with PyHessian. If not, see <http://www.gnu.org/licenses/>.
#*
from __future__ import print_function
import os
import sys
import json
import time
import random
import argparse
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torch.autograd import Variable
from torchvision import datasets, transforms
from pyhessian import hessian, hessian_input
from process.pyhessian_density_plot import get_esd_plot
from utils.pruner import *
import torchvision.models as models
from models.ResNet_cifar import resnet20, resnet20_small
from models.ResNet import resnet18, resnet18_small
from models.WideResNet_cifar import WideResNet, WideResNet_small
from advertorch.utils import NormalizeByChannelMeanStd
from datasets import pure_cifar10_dataloaders, pure_cifar100_dataloaders, pure_imagenet_dataloader
from models.ResNet_img import resnet50_small
# Settings
parser = argparse.ArgumentParser(description='PyTorch pyhessian analysis')
parser.add_argument('--data', type=str, default='../data', help='location of the data corpus')
parser.add_argument('--dataset', type=str, default='cifar10', help='dataset')
parser.add_argument('--batch_size', type=int, default=200, help='batch size')
parser.add_argument('--batch_number', type=int, default=1, help='Data number = batch_size*batch_number')
parser.add_argument('--arch', type=str, default='resnet20', help='architecture')
parser.add_argument('--pretrained', type=str, default=None, help='pretrained weight')
parser.add_argument('--output_file', type=str, default='test.pt', help="the name of output file")
parser.add_argument('--mode', type=str, default='weight')
parser.add_argument('--workers', type=int, default=4, help='number of workers in dataloader')
parser.add_argument('--seed', type=int, default=1, help='random seed (default: 1)')
parser.add_argument('--gpu', type=int, default=0, help='GPU ID')
args = parser.parse_args()
torch.cuda.set_device(args.gpu)
# set random seed to reproduce the work
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
np.random.seed(args.seed)
random.seed(args.seed)
torch.backends.cudnn.deterministic = True
for arg in vars(args):
print(arg, getattr(args, arg))
if os.path.isfile(args.output_file):
result_dict = torch.load(args.output_file)
else:
result_dict = {}
#############################################################################
######################### Get the hessian data ############################
#############################################################################
if args.dataset == 'cifar10':
print('Dataset = CIFAR10')
classes = 10
normalization = NormalizeByChannelMeanStd(
mean=[0.4914, 0.4822, 0.4465], std=[0.2470, 0.2435, 0.2616])
train_loader = pure_cifar10_dataloaders(args.batch_size, args.data)
elif args.dataset == 'cifar100':
print('Dataset = CIFAR100')
classes = 100
normalization = NormalizeByChannelMeanStd(
mean=[0.5071, 0.4866, 0.4409], std=[0.2673, 0.2564, 0.2762])
train_loader = pure_cifar100_dataloaders(args.batch_size, args.data)
elif args.dataset == 'imagenet':
print('Dataset = ImageNet')
train_loader = pure_imagenet_dataloader(args, use_normalize=True)
else:
raise ValueError(('unsupport dataset'))
if args.batch_number == 1:
for inputs, labels in train_loader:
hessian_dataloader = (inputs, labels)
break
else:
hessian_dataloader = []
for i, (inputs, labels) in enumerate(train_loader):
hessian_dataloader.append((inputs, labels))
if i == args.batch_number - 1:
break
#############################################################################
############################## Get the model ##############################
#############################################################################
if 'cifar' in args.dataset:
if args.arch == 'resnet20s':
print('build model resnet20_cifar')
model = resnet20(number_class=classes)
elif args.arch == 'resnet18':
print('build model resnet18')
model = resnet18(num_classes=classes, imagenet=False)
elif args.arch == 'wideresnet':
print('build model: wideresnet-28-10')
model = WideResNet(depth=28, num_classes=classes, widen_factor=10, dropRate=0.0)
elif args.arch == 'resnet20s_small':
print('build model resnet20_cifar_small')
model = resnet20_small(number_class=classes)
elif args.arch == 'resnet18_small':
print('build model resnet18_small')
model = resnet18_small(num_classes=classes, imagenet=False)
elif args.arch == 'wideresnet_small':
print('build model: wideresnet-28-10_small')
model = WideResNet_small(depth=28, num_classes=classes, widen_factor=10, dropRate=0.0)
else:
raise ValueError('unknow model')
model.normalize = normalization
else:
if 'small' in args.arch:
print('build resnet50 small')
model = resnet50_small()
else:
print('build resnet50')
model = models.resnet50(pretrained=False)
model.cuda()
if args.pretrained:
print('===> loading weight from {} <==='.format(args.pretrained))
pretrained_weight = torch.load(args.pretrained, map_location = torch.device('cuda:'+str(args.gpu)))
if 'state_dict' in pretrained_weight:
pretrained_weight = pretrained_weight['state_dict']
sparse_mask = extract_mask(pretrained_weight)
if len(sparse_mask) > 0:
prune_model_custom(model, sparse_mask)
model.load_state_dict(pretrained_weight)
check_sparsity(model)
#############################################################################
########################## Begin the computation ##########################
#############################################################################
if 'weight' in args.mode:
print('hessian analysis respect to weight')
criterion = nn.CrossEntropyLoss()
model.eval()
if args.batch_number == 1:
hessian_comp = hessian(model, criterion, data=hessian_dataloader)
else:
hessian_comp = hessian(model, criterion, dataloader=hessian_dataloader)
print('********** finish data loading and begin Hessian computation **********')
start = time.time()
top_eigenvalues, _ = hessian_comp.eigenvalues()
end = time.time()
trace = hessian_comp.trace()
end2 = time.time()
print(end - start)
print(end2 - end)
print('\n***Top Eigenvalues: ', top_eigenvalues)
print('\n***Trace: ', np.mean(trace))
result_dict['eigenvalues'] = top_eigenvalues
result_dict['trace'] = trace
torch.save(result_dict, args.output_file)
##########################################################################################
############## Begin the computation hessian respect to input ##########################
##########################################################################################
if 'input' in args.mode:
print('hessian analysis respect to input')
criterion = nn.CrossEntropyLoss()
model.eval()
if args.batch_number == 1:
hessian_comp_input = hessian_input(model, criterion, data=hessian_dataloader)
else:
hessian_comp_input = hessian_input(model, criterion, data=hessian_dataloader[0])
print('********** finish data loading and begin Hessian computation **********')
start = time.time()
top_eigenvalues_input, _ = hessian_comp_input.eigenvalues()
end = time.time()
trace_input = hessian_comp_input.trace()
end2 = time.time()
print('\n***Top Eigenvalues (input): ', top_eigenvalues_input)
print('\n***Trace (input): ', np.mean(trace_input))
result_dict['eigenvalues_input'] = top_eigenvalues_input
result_dict['trace_input'] = trace_input
print(end - start)
print(end2 - end)
torch.save(result_dict, args.output_file)