forked from eliberis/uNAS
-
Notifications
You must be signed in to change notification settings - Fork 0
/
metrics_ntk_v2.py
250 lines (210 loc) · 8.94 KB
/
metrics_ntk_v2.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
'''
Copyright (C) 2010-2021 Alibaba Group Holding Limited.
This file is modified from:
https://github.com/VITA-Group/TENAS
'''
import ray
#################### TEGNAS testntk ####################
import numpy as np
import torch
import torch.nn as nn
import tf2onnx
import onnx
import onnx2torch
import pickle
import pdb
import gc
import os
import time
#################### GPU ####################
#os.environ['CUDA_LAUNCH_BLOCKING'] = "1"
ray.init(num_cpus=2, num_gpus=1)
@ray.remote(num_gpus=1, max_calls=1)
def get_ntk(num_batch=1, networks_num=3):
save_path = './tmp/metrics'
num_batch = num_batch
timestamp = ''
networks_num = networks_num
num_classes = 10
ntks = -1
# return timestamp, num_batch
def wait_input(save_path):
input_finish_info_path = f'{save_path}/input_finish_info.npz'
while not os.path.isfile(input_finish_info_path):
time.sleep(5)
print (f'wait for input_finish_info')
print (f'find input_finish_info')
time.sleep(5)
#load input_finish_info and del
input_finish_info = np.load(input_finish_info_path)
num_batch = int(input_finish_info['num_batch'])
num_classes = int(input_finish_info['num_classes'])
timestamp = str(input_finish_info['timestamp'])
os.remove(input_finish_info_path)
return num_batch, num_classes, timestamp
# return (train_loader, val_loader)
def load_dataset(save_path, num_batch):
#################### dataset ####################
'''
#load loader
loader_save_path = f'{save_path}/loader.npz'
loader = np.load(loader_save_path)
train_input = loader['train_input']
train_target = loader['train_target']
val_input = loader['val_input']
val_target = loader['val_target']
# for get ntk loader input
train_loader = []
val_loader = []
for i in range(num_batch):
train_loader.append((train_input,train_target))
val_loader.append((val_input,val_target))
'''
# train_loader.pickle/val_loader.pickle
train_loader_save_path = f'{save_path}/train_loader.pickle'
with open(train_loader_save_path, 'rb') as f:
train_loader = pickle.load(f)
val_loader_save_path = f'{save_path}/val_loader.pickle'
with open(val_loader_save_path, 'rb') as f:
val_loader = pickle.load(f)
#clear the parameter
#del train_input, train_target
#del val_input, val_target
gc.collect()
return train_loader, val_loader
# return model (pytorch)
def transfer_init_model(save_path):
#################### model ####################
# load onnx_model
onnx_model_path = onnx_model_path = f'{save_path}/model.onnx'
onnx_model = onnx.load(onnx_model_path)
# onnx2torch
torch_model = onnx2torch.convert(onnx_model)
# Model class must be defined somewhere
torch_model.eval()
# torch_model 參數初始化
def kaiming_normal_fanin_init(m):
if isinstance(m, nn.Conv2d) or isinstance(m, nn.Linear):
nn.init.kaiming_normal_(m.weight, mode='fan_in', nonlinearity='relu')
if hasattr(m, 'bias') and m.bias is not None:
nn.init.zeros_(m.bias)
elif isinstance(m, nn.BatchNorm2d):
nn.init.ones_(m.weight.data)
nn.init.constant_(m.bias.data, 0.0)
def kaiming_normal_fanout_init(m):
if isinstance(m, nn.Conv2d) or isinstance(m, nn.Linear):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
if hasattr(m, 'bias') and m.bias is not None:
nn.init.zeros_(m.bias)
elif isinstance(m, nn.BatchNorm2d):
nn.init.ones_(m.weight.data)
nn.init.constant_(m.bias.data, 0.0)
def init_model(model, method='kaiming_norm_fanout'):
if method == 'kaiming_norm_fanin':
model.apply(kaiming_normal_fanin_init)
elif method == 'kaiming_norm_fanout':
model.apply(kaiming_normal_fanout_init)
return model
model = init_model (torch_model)
#clear the parameter
del onnx_model
del torch_model
gc.collect()
return model
# return (conds)
def get_ntk_n(loader, networks, train_mode=True, num_batch=1, num_classes=10):
print (f'num_batch={num_batch}')
print (f'num_classes={num_classes}')
#################### ntk ####################
device = torch.cuda.current_device()
# if recalbn > 0:
# network = recal_bn(network, xloader, recalbn, device)
# if network_2 is not None:
# network_2 = recal_bn(network_2, xloader, recalbn, device)
ntks = []
for network in networks:
if train_mode:
network.train()
else:
network.eval()
######
# 建立grads list,將裡面有數量為networks list長度的空list
grads = [[] for _ in range(len(networks))]
# xloader 內有 inputs和targets
for i, (inputs, targets) in enumerate(loader):
# num_batch 預設為-1
if num_batch > 0 and i >= num_batch: break
# numpy to pytorch tensor
# torch.cuda.DoubleTensor to torch.cuda.FloatTensor
inputs = torch.from_numpy(inputs).float()
targets = torch.from_numpy(targets)
# 將inputs, targets放入gpu
print (f'inputs={inputs.shape}')
inputs = inputs.to(device, non_blocking=True)
# 對networks list內的每個network
for net_idx, network in enumerate(networks):
# 將network(weight)放入gpu
network.to(device)
# 將network的梯度歸零
network.zero_grad()
# 會將梯度疊加給inputs_
inputs_ = inputs.clone().to(device, non_blocking=True)
# logit 是inputs_作為輸入的netowrk輸出
logit = network(inputs_)
# 若logit 是tuple的話
if isinstance(logit, tuple):
# 則logit 將變成logit tuple第二個元素
logit = logit[1] # 201 networks: return features and logits
# 將每個inputs_送進network中,並將梯度輸出放在logit list內
for _idx in range(len(inputs_)):
# 對於在inputs_中的每個input,傳入和輸出一樣shape且全為1的矩陣,可得到所有子結點的梯度
logit[_idx:_idx+1].backward(torch.ones_like(logit[_idx:_idx+1]), retain_graph=True)
# 建立梯度list
grad = []
# 對所有netowrk的參數
for name, W in network.named_parameters():
# 將權重梯度append進grad中
if 'weight' in name and W.grad is not None:
grad.append(W.grad.view(-1).detach())
# 再將grad放進grads list中
grads[net_idx].append(torch.cat(grad, -1))
# 將network梯度歸零
network.zero_grad()
# 清空cache
torch.cuda.empty_cache()
#pdb.set_trace()
######
#
grads = [torch.stack(_grads, 0) for _grads in grads]
ntks = [torch.einsum('nc,mc->nm', [_grads, _grads]) for _grads in grads]
conds = []
for ntk in ntks:
eigenvalues, _ = torch.symeig(ntk) # ascending
conds.append(np.nan_to_num((eigenvalues[-1] / eigenvalues[0]).item(), copy=True, nan=100000.0))
return conds
def save_ntks(save_path, ntks):
# save ntks as ntks.npz
ntks_save_path = f'{save_path}/ntks.npz'
np.savez(ntks_save_path, ntks=ntks)
return
# wait input
num_batch, num_classes, timestamp = wait_input(save_path)
# loaddataset
train_loader, val_loader = load_dataset(save_path, num_batch)
# transfer and init model
networks = []
for i in range(networks_num):
# transfer and init model
torch_model = transfer_init_model(save_path)
networks.append(torch_model)
# get ntk_n
ntks = get_ntk_n(loader=train_loader, networks=networks, train_mode=True, num_batch=num_batch, num_classes=num_classes)
# save_ntks
save_ntks(save_path, ntks)
# save metrics_finish_info
timestamp = "{:}".format(time.strftime('%h-%d-%C_%H-%M-%s', time.localtime(time.time())))
metrics_finish_info_path = f'{save_path}/metrics_finish_info.npz'
np.savez(metrics_finish_info_path, timestamp=timestamp)
return
while True:
ntks = ray.get(get_ntk.remote(networks_num=3, num_batch=1))