-
Notifications
You must be signed in to change notification settings - Fork 8
Expand file tree
/
Copy pathmy_exp_config_old.py
More file actions
172 lines (161 loc) · 5.42 KB
/
my_exp_config_old.py
File metadata and controls
172 lines (161 loc) · 5.42 KB
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
def get_caffe_prms_old(nwPrms, lrPrms, finePrms=None,
isScratch=True, deviceId=1,
runNum=0, resumeIter=0):
caffePrms = edict()
caffePrms.deviceId = deviceId
caffePrms.isScratch = isScratch
caffePrms.nwPrms = copy.deepcopy(nwPrms)
caffePrms.lrPrms = copy.deepcopy(lrPrms)
caffePrms.finePrms = copy.deepcopy(finePrms)
caffePrms.resumeIter = resumeIter
expStr = nwPrms.expStr + '/' + lrPrms.expStr
if finePrms is not None:
expStr = expStr + '/' + finePrms.expStr
if runNum > 0:
expStr = expStr + '_run%d' % runNum
caffePrms['expStr'] = expStr
caffePrms['solver'] = lrPrms.solver
return caffePrms
##
# Parameters required to specify the n/w architecture
def get_nw_prms(isHashStr=False, **kwargs):
dArgs = edict()
dArgs.netName = 'alexnet'
dArgs.concatLayer = 'fc6'
dArgs.concatDrop = False
dArgs.contextPad = 0
dArgs.imSz = 227
dArgs.imgntMean = True
dArgs.maxJitter = 0
dArgs.randCrop = False
dArgs.lossWeight = 1.0
dArgs.multiLossProto = None
dArgs.ptchStreamNum = 256
dArgs.poseStreamNum = 256
dArgs.isGray = False
dArgs.isPythonLayer = False
dArgs.extraFc = None
dArgs.numFc5 = None
dArgs.numConv4 = None
dArgs.numCommonFc = None
dArgs.lrAbove = None
dArgs = mpu.get_defaults(kwargs, dArgs)
if dArgs.numFc5 is not None:
assert(dArgs.concatLayer=='fc5')
expStr = 'net-%s_cnct-%s_cnctDrp%d_contPad%d_imSz%d_imgntMean%d_jit%d'\
%(dArgs.netName, dArgs.concatLayer, dArgs.concatDrop,
dArgs.contextPad,
dArgs.imSz, dArgs.imgntMean, dArgs.maxJitter)
if dArgs.numFc5 is not None:
expStr = '%s_numFc5-%d' % (expStr, dArgs.numFc5)
if dArgs.numConv4 is not None:
expStr = '%s_numConv4-%d' % (expStr, dArgs.numConv4)
if dArgs.numCommonFc is not None:
expStr = '%s_numCommonFc-%d' % (expStr, dArgs.numCommonFc)
if dArgs.randCrop:
expStr = '%s_randCrp%d' % (expStr, dArgs.randCrop)
if not(dArgs.lossWeight==1.0):
if type(dArgs.lossWeight)== list:
lStr = ''
for i,l in enumerate(dArgs.lossWeight):
lStr = lStr + 'lw%d-%.1f_' % (i,l)
lStr = lStr[0:-1]
print lStr
expStr = '%s_%s' % (expStr, lStr)
else:
assert isinstance(dArgs.lossWeight, (int, long, float))
expStr = '%s_lw%.1f' % (expStr, dArgs.lossWeight)
if dArgs.multiLossProto is not None:
expStr = '%s_mlpr%s-posn%d-ptsn%d' % (expStr,
dArgs.multiLossProto, dArgs.poseStreamNum, dArgs.ptchStreamNum)
if dArgs.isGray:
expStr = '%s_grayIm' % expStr
if dArgs.isPythonLayer:
expStr = '%s_pylayers' % expStr
if dArgs.extraFc is not None:
expStr = '%s_extraFc%d' % (expStr, dArgs.extraFc)
if dArgs.lrAbove is not None:
expStr = '%s_lrAbove-%s' % (expStr, dArgs.lrAbove)
if not isHashStr:
dArgs.expStr = expStr
else:
dArgs.expStr = 'nwPrms-%s' % ou.hash_dict_str(dArgs)
return dArgs
##
# Parameters that specify the learning
def get_lr_prms(isHashStr=False, **kwargs):
dArgs = edict()
dArgs.batchsize = 128
dArgs.stepsize = 20000
dArgs.base_lr = 0.001
dArgs.max_iter = 250000
dArgs.gamma = 0.5
dArgs.weight_decay = 0.0005
dArgs.clip_gradients = -1
dArgs.debug_info = False
dArgs = mpu.get_defaults(kwargs, dArgs)
#Make the solver
debugStr = '%s' % dArgs.debug_info
debugStr = debugStr.lower()
del dArgs['debug_info']
solArgs = edict({'test_iter': 100, 'test_interval': 1000,
'snapshot': 2000,
'debug_info': debugStr})
#print dArgs.keys()
expStr = 'batchSz%d_stepSz%.0e_blr%.5f_mxItr%.1e_gamma%.2f_wdecay%.6f'\
% (dArgs.batchsize, dArgs.stepsize, dArgs.base_lr,
dArgs.max_iter, dArgs.gamma, dArgs.weight_decay)
if not(dArgs.clip_gradients==-1):
expStr = '%s_gradClip%.1f' % (expStr, dArgs.clip_gradients)
if not isHashStr:
dArgs.expStr = expStr
else:
dArgs.expStr = 'lrPrms-%s' % ou.hash_dict_str(dArgs)
for k in dArgs.keys():
if k in ['batchsize', 'expStr']:
continue
solArgs[k] = copy.deepcopy(dArgs[k])
dArgs.solver = mpu.make_solver(**solArgs)
return dArgs
##
# Parameters for fine-tuning
def get_finetune_prms(isHashStr=False, **kwargs):
'''
sourceModelIter: The number of model iterations of the source model to consider
fine_max_iter : The maximum iterations to which the target model should be trained.
lrAbove : If learning is to be performed some layer.
fine_base_lr : The base learning rate for finetuning.
fineRunNum : The run num for the finetuning.
fineNumData : The amount of data to be used for the finetuning.
fineMaxLayer : The maximum layer of the source n/w that should be considered.
'''
dArgs = edict()
dArgs.base_lr = 0.001
dArgs.runNum = 1
dArgs.maxLayer = None
dArgs.lrAbove = None
dArgs.dataset = 'sun'
dArgs.maxIter = 40000
dArgs.extraFc = False
dArgs.extraFcDrop = False
dArgs.sourceModelIter = 60000
dArgs = mpu.get_defaults(kwargs, dArgs)
return dArgs
def get_caffe_prms(nwPrms, lrPrms, finePrms=None,
isScratch=True, deviceId=1,
runNum=0, resumeIter=0):
caffePrms = edict()
caffePrms.deviceId = deviceId
caffePrms.isScratch = isScratch
caffePrms.nwPrms = copy.deepcopy(nwPrms)
caffePrms.lrPrms = copy.deepcopy(lrPrms)
caffePrms.finePrms = copy.deepcopy(finePrms)
caffePrms.resumeIter = resumeIter
expStr = nwPrms.expStr + '/' + lrPrms.expStr
if finePrms is not None:
expStr = expStr + '/' + finePrms.expStr
if runNum > 0:
expStr = expStr + '_run%d' % runNum
caffePrms['expStr'] = expStr
caffePrms['solver'] = lrPrms.solver
return caffePrms