-
Notifications
You must be signed in to change notification settings - Fork 0
/
mask.py
224 lines (195 loc) · 8.3 KB
/
mask.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
import torch
import numpy as np
import pickle
import utils
class mask_vgg_16_bn:
def __init__(self, model=None, job_dir='',device=None):
self.model = model
self.mask = {}
self.job_dir=job_dir
self.device = device
def layer_mask(self, cov_id, resume=None, param_per_cov=4, rank = None, type = 1, arch="vgg_16_bn"):
params = self.model.parameters()
prefix = "rank_conv/"+arch+"/rank_conv"
subfix = ".npy"
if resume:
with open(resume, 'rb') as f:
self.mask = pickle.load(f)
else:
resume='mask/mask_vgg_16_bn'
self.param_per_cov=param_per_cov
for index, item in enumerate(params):
if index == (cov_id + 1) * param_per_cov:
break
if index == cov_id * param_per_cov:
f, c, w, h = item.size()
if rank is None:
rank = np.load(prefix + str(cov_id) + subfix)
utils.logger.info('loading '+prefix + str(cov_id) + subfix)
if type == 1:
inds = torch.zeros(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 1
self.mask[index] = inds
else :
inds = torch.ones(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 0
self.mask[index] = inds
item.data = item.data * self.mask[index]
if index > cov_id * param_per_cov and index < (cov_id + 1) * param_per_cov :
self.mask[index] = torch.squeeze(inds)
item.data = item.data * self.mask[index]
with open(resume, "wb") as f:
pickle.dump(self.mask, f)
def grad_mask(self, cov_id):
params = self.model.parameters()
mask_keys = self.mask.keys()
for index, item in enumerate(params):
if index < cov_id * self.param_per_cov:
continue
if index in mask_keys:
item.data = item.data * self.mask[index]
class mask_resnet_56:
def __init__(self, model=None, job_dir='',device=None):
self.model = model
self.mask = {}
self.job_dir=job_dir
self.device = device
def layer_mask(self, cov_id, resume=None, param_per_cov=3, rank = None, type = 1, arch="resnet_56"):
params = self.model.parameters()
prefix = "rank_conv/"+arch+"/rank_conv"
subfix = ".npy"
if resume:
with open(resume, 'rb') as f:
self.mask = pickle.load(f)
else:
resume='mask/mask_resnet_56'
self.param_per_cov=param_per_cov
for index, item in enumerate(params):
if index == (cov_id + 1)*param_per_cov:
break
if index == cov_id * param_per_cov:
f, c, w, h = item.size()
if rank is None:
rank = np.load(prefix + str(cov_id) + subfix)
utils.logger.info('loading '+prefix + str(cov_id) + subfix)
if type == 1:
inds = torch.zeros(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 1
self.mask[index] = inds
else :
inds = torch.ones(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 0
self.mask[index] = inds
item.data = item.data * self.mask[index]
elif index > cov_id*param_per_cov and index < (cov_id + 1)*param_per_cov:
self.mask[index] = torch.squeeze(inds)
item.data = item.data * self.mask[index].to(self.device)
with open(resume, "wb") as f:
pickle.dump(self.mask, f)
def grad_mask(self, cov_id):
if len(self.mask.keys()) < 1 : return
params = self.model.parameters()
mask_keys = self.mask.keys()
for index, item in enumerate(params):
if index == (cov_id + 1)*self.param_per_cov:
break
if index in mask_keys:
item.data = item.data * self.mask[index].to(self.device)#prune certain weight
class mask_googlenet:
def __init__(self, model=None, job_dir='',device=None):
self.model = model
self.mask = {}
self.job_dir=job_dir
self.device = device
def layer_mask(self, cov_id, resume=None, param_per_cov=4, rank = None, type = 1, arch="googlenet"):
params = self.model.parameters()
prefix = "rank_conv/"+arch+"/rank_conv"
subfix = ".npy"
if resume:
with open(resume, 'rb') as f:
self.mask = pickle.load(f)
else:
resume='mask/mask_googlenet'
self.param_per_cov=param_per_cov
for index, item in enumerate(params):
if index == (cov_id + 1) * param_per_cov:
break
if index == cov_id * param_per_cov:
f, c, w, h = item.size()
if rank is None:
rank = np.load(prefix + str(cov_id) + subfix)
utils.logger.info('loading '+prefix + str(cov_id) + subfix)
if type == 1:
inds = torch.zeros(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 1
self.mask[index] = inds
else :
inds = torch.ones(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 0
self.mask[index] = inds
item.data = item.data * self.mask[index]
if index > cov_id * param_per_cov and index < (cov_id + 1) * param_per_cov :
self.mask[index] = torch.squeeze(inds)
item.data = item.data * self.mask[index]
with open(resume, "wb") as f:
pickle.dump(self.mask, f)
def grad_mask(self, cov_id):
if len(self.mask.keys()) < 1 : return
params = self.model.parameters()
mask_keys = self.mask.keys()
for index, item in enumerate(params):
if index == cov_id * self.param_per_cov + 4:
break
if index in mask_keys:
item.data = item.data * self.mask[index].to(self.device)#prune certain weight
class mask_resnet_50:
def __init__(self, model=None, job_dir='',device=None):
self.model = model
self.mask = {}
self.job_dir=job_dir
self.device = device
def layer_mask(self, cov_id, resume=None, param_per_cov=3, rank = None, type = 1, arch="resnet_50"):
assert(rank is not None)
params = self.model.parameters()
if resume:
with open(resume, 'rb') as f:
self.mask = pickle.load(f)
else:
resume='mask/mask_resnet_50'
self.param_per_cov=param_per_cov
for index, item in enumerate(params):
if index == (cov_id + 1) * param_per_cov:
break
if index == cov_id * param_per_cov:
f, c, w, h = item.size()
if type == 1:
inds = torch.zeros(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 1
self.mask[index] = inds
else :
inds = torch.ones(f, 1, 1, 1).to(self.device)
for i in rank:
inds[i, 0, 0, 0] = 0
self.mask[index] = inds
item.data = item.data * self.mask[index]
elif index > cov_id * param_per_cov and index < (cov_id + 1)* param_per_cov:
self.mask[index] = torch.squeeze(inds)
item.data = item.data * self.mask[index]
with open(resume, "wb") as f:
pickle.dump(self.mask, f)
def grad_mask(self, cov_id):
if len(self.mask.keys()) < 1 : return
params = self.model.parameters()
mask_keys = self.mask.keys()
for index, item in enumerate(params):
if index == (cov_id + 1)*self.param_per_cov:
break
if index in mask_keys:
item.data = item.data * self.mask[index].to(self.device)