-
Notifications
You must be signed in to change notification settings - Fork 1
/
function.py
196 lines (158 loc) · 5.83 KB
/
function.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
import torch
# import pdb
import matplotlib.pyplot as plt
import numpy as np
import time
def updateTime(time2Update,timeStart):
timePassed = time.time()-timeStart
time2Update += timePassed
return time2Update
def movingAvg(oldAvg,beta,newData):
newAvg = beta*oldAvg + (1-beta)*newData
return newAvg
def selectValData(ValLoader,batchSize):
numOfData = len(ValLoader.dataset)
randLoc = torch.randperm(numOfData)[:batchSize]
valData = torch.zeros(batchSize,3,224,224)
valTarget = torch.zeros(batchSize,dtype=torch.long)
for it in range(batchSize):
valData[it] = ValLoader.dataset[randLoc[it]][0]
valTarget[it] = ValLoader.dataset[randLoc[it]][1]
return valData,valTarget
def createDataset(save,**kwargs):
with torch.no_grad():
for key,value in kwargs.items():
mean = increaseDim(value)
kwargs = {key:mean}
save.createDataset(**kwargs)
def saveDataset(save,**kwargs):
with torch.no_grad():
for key,value in kwargs.items():
mean = increaseDim(value)
kwargs = {key:mean}
save.variable2Write(**kwargs)
def saveNetworkParam(save,net):
with torch.no_grad():
for name,param in net.named_parameters():
tempName = name.split('.')
if tempName[0][-1] !='B':
mean = calculateMeanStd(param.data)
kwargs = {name:mean}
save.variable2Write(**kwargs)
mean = calculateMeanStd(param.grad.data)
kwargs = {name+'.grad':mean}
save.variable2Write(**kwargs)
def createNetworkParam(save,net):
with torch.no_grad():
for name,param in net.named_parameters():
tempName = name.split('.')
if tempName[0][-1] !='B':
mean = calculateMeanStd(param.data)
kwargs = {name:mean}
save.createDataset(**kwargs)
mean = calculateMeanStd(param.grad.data)
kwargs = {name+'.grad':mean}
save.createDataset(**kwargs)
def calculateMeanStd(data):
with torch.no_grad():
if len(data.shape) == 1:
meanData = increaseDim(data)
elif len(data.shape)== 2:
meanData = torch.mean(data,dim=0)
elif len(data.shape)== 4:
meanData = torch.mean(data,dim=1)
meanData = increaseDim(meanData)
return meanData
def increaseDim(data1):
with torch.no_grad():
return data1.unsqueeze(0)
def adjust_lr(optimizer,iter,epoch,learRate,power):
lr = learRate*(1 - iter/epoch)**power
for param_group in optimizer.param_groups:
param_group['lr'] = lr
def accuracy(output, target, topk=(1,)):
with torch.no_grad():
maxk = max(topk)
batchSize = target.size(0)
_, pred = output.topk(maxk, 1, True, True)
pred = pred.t()
correct = pred.eq(target.view(1, -1).expand_as(pred))
res = []
for k in topk:
correct_k = correct[:k].view(-1).float().sum(0, keepdim=True)
res.append(correct_k.mul_(100.0 / batchSize))
return res
def initKernels(fileName,net):
data = torch.load(fileName)
oldNet = data['net']
with torch.no_grad():
for data1,data2 in zip(oldNet.named_parameters(),net.named_parameters()):
tempName = data1[0].split('.')
if tempName[0][-1] !='B':
data2[1] .data = data1[1].data
return net
def initWeights(net):
with torch.no_grad():
for data1 in net.named_parameters():
torch.nn.init.normal_(data1[1],mean = 0,std= 0.01)
return net
def cycleParam(startVals,stopVals,percents,types,totalNum):
# import pdb;pdb.set_trace()
param = []
for startVal,stopVal,percent,typeR in zip(startVals,stopVals,percents,types):
tempNum = int((totalNum*percent/100))
if typeR == 'lin':
param[-1:] = torch.linspace(startVal,stopVal,tempNum)
elif typeR == 'cos':
param[-1:] = stopVal+(startVal-stopVal)*torch.tensor(np.cos(2*np.pi*1/(tempNum)*np.linspace(0,tempNum/2,tempNum))+1)/2
param = torch.tensor(param)
return param.reshape((1,-1))
# def cycleParam(startVals,stopVals,percents,totalNum):
# param = []
# for startVal,stopVal,percent in zip(startVals,stopVals,percents):
# tempNum = int((totalNum*percent/100))
# param[-1:] = torch.linspace(startVal,stopVal,tempNum)
# params = torch.tensor(param)
# return params.reshape((1,-1))
def logLinearLR(startVal,stopVal,itNum):
logStart = torch.log10(torch.tensor(startVal).type(torch.float))
logStop = torch.log10(torch.tensor(stopVal).type(torch.float))
tempLr = torch.logspace(logStart,logStop,itNum)
return tempLr
def linearLR(startVal,stopVal,itNum):
tempLr = torch.linspace(startVal,stopVal,itNum)
return tempLr
def loadCheckpoint(fileName):
data = torch.load(fileName)
epoch = data['epoch']
net = data['net']
optim = data['optim']
loss = data['loss']
top1 = data['top1']
top5 = data['top5']
return epoch,net,optim,loss,top1,top5
def saveCheckPoint(fileName,epoch,net,optimizer,loss,top1,top5,isBest):
data={}
data['epoch'] = epoch
data['net'] = net
data['loss'] = loss
data['top1'] = top1
data['top5'] = top5
data['optim'] = optimizer
if isBest:
print('data is written to disk as best')
torch.save(data,'BestData.pth')
torch.save(data,fileName)
else:
print('data is written to disk')
torch.save(data,fileName)
if __name__=="__main__":
startVals =[0.7,5,7]
stopVals = [5 ,7,0.07]
percents = [50,10,40]
typeR = ['lin','lin','cos']
param = cycleParam(startVals,stopVals,percents,typeR,500)
print(param.shape)
plt.figure()
plt.plot(np.array(param).reshape((-1)))
plt.show()