ADAPT-pNC/configuration.py at master · KIT-Neuromorphic-Computing/ADAPT-pNC · GitHub

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
import argparse

parser = argparse.ArgumentParser(
    # prog = 'Printed Neuromorphic Circuits with Learnable Filters',
    description='Training for Printed Neuromorphic Circuits with Learnable Filters')

# printing-related hyperparameters for pNNs
parser.add_argument('--gmin',                  type=float,     default=0.01,
                    help='minimal printable conductance value')
parser.add_argument('--gmax',                  type=float,     default=10.,
                    help='maximal printable conductance value')
parser.add_argument('--T',                     type=float,
                    default=0.1,                   help='measuring threshold')
parser.add_argument('--m',                     type=float,
                    default=0.3,                   help='measuring margin')
# learnable tanh circuits
parser.add_argument('--ACT_R1n',               type=float,
                    default=12.7865,               help='resistance in nonlinear circuit')
parser.add_argument('--ACT_R2n',               type=float,     default=-
                    3.1871,               help='resistance in nonlinear circuit')
parser.add_argument('--ACT_W1n',               type=float,
                    default=-10.4537,              help='width of the transistor 1')
parser.add_argument('--ACT_L1n',               type=float,
                    default=10.5460,               help='length of the transistor 1')
parser.add_argument('--ACT_W2n',               type=float,
                    default=-5.8496,               help='width of the transistor 2')
parser.add_argument('--ACT_L2n',               type=float,
                    default=4.2337,                help='length of the transistor 2')
# learnable sigmoid circuits
parser.add_argument('--S_R1n',                 type=float,     default=-
                    11.7481,              help='resistance in nonlinear circuit')
parser.add_argument('--S_R2n',                 type=float,     default=-
                    0.3905,               help='resistance in nonlinear circuit')
parser.add_argument('--S_W1n',                 type=float,
                    default=-9.1690,               help='width of the transistor')
parser.add_argument('--S_L1n',                 type=float,
                    default=0.3913,                help='length of the transistor')
parser.add_argument('--S_W2n',                 type=float,
                    default=7.2004,                help='width of the transistor')
parser.add_argument('--S_L2n',                 type=float,
                    default=15.5558,               help='length of the transistor')
# learnable pReLU circuits
parser.add_argument('--ReLU_RHn',              type=float,     default=-
                    1.9557,               help='resistance in nonlinear circuit')
parser.add_argument('--ReLU_RLn',              type=float,     default=-
                    20.1581,              help='resistance in nonlinear circuit')
parser.add_argument('--ReLU_RDn',              type=float,
                    default=16.6486,               help='resistance in nonlinear circuit')
parser.add_argument('--ReLU_RBn',              type=float,     default=-
                    14.3472,              help='resistance in nonlinear circuit')
parser.add_argument('--ReLU_Wn',               type=float,
                    default=16.6550,               help='width of the transistor')
parser.add_argument('--ReLU_Ln',               type=float,
                    default=-21.7591,              help='length of the transistor')
# learnable hard sigmoid circuits
parser.add_argument('--HS_Rn',                 type=float,     default=-
                    0.3766,               help='resistance in nonlinear circuit')
parser.add_argument('--HS_Wn',                 type=float,
                    default=0.5192,                help='width of the transistor')
parser.add_argument('--HS_Ln',                 type=float,
                    default=4.6293,                help='length of the transistor')
# learnable negative weight circuits
parser.add_argument('--NEG_R1n',               type=float,     default=-
                    7.0846,               help='resistance in nonlinear circuit')
parser.add_argument('--NEG_R2n',               type=float,     default=-
                    2.5695,               help='resistance in nonlinear circuit')
parser.add_argument('--NEG_R3n',               type=float,
                    default=16.1542,               help='resistance in nonlinear circuit')
parser.add_argument('--NEG_W1n',               type=float,
                    default=13.8696,               help='width of the transistor')
parser.add_argument('--NEG_L1n',               type=float,
                    default=2.7635,                help='length of the transistor')
parser.add_argument('--NEG_W2n',               type=float,
                    default=1.2674,                help='width of the transistor')
parser.add_argument('--NEG_L2n',               type=float,
                    default=-10.4803,              help='length of the transistor')
parser.add_argument('--NEG_W3n',               type=float,
                    default=-6.4386,               help='width of the transistor')
parser.add_argument('--NEG_L3n',               type=float,
                    default=5.1005,                help='length of the transistor')

# machine-learning-related hyperparameters
# dataset-related
parser.add_argument('--task',                  type=str,       default='temporized',
                    help='train normal pNN or split manufacturing, or temporal information')
parser.add_argument('--DS_VAR',               type=str,
                    default='none',                help='variation of dataset')
parser.add_argument('--N_Channel',               type=int,
                    default=2,                help='number of layers')
parser.add_argument('--N_feature',               type=int,
                    default=None,                help='number of LF in group')

parser.add_argument('--DATASET',               type=int,
                    default=0,                     help='index of training dataset')
parser.add_argument('--DataPath',              type=str,
                    default='./dataset',           help='path to dataset')
parser.add_argument('--augment',              type=bool,
                    default=False,                help='if needed data augmentation')
# data augmentation
parser.add_argument('--InputNoise',            type=float,
                    default=0.,                    help='noise of input signal')
parser.add_argument('--max_drift',            type=float,
                    default=0.2,                    help='noise of input signal')
parser.add_argument('--scale',            type=float,
                    default=0.02,                    help='noise of input signal')
parser.add_argument('--n_speed_change',            type=int,
                    default=5,                    help='noise of input signal')
parser.add_argument('--max_speed_ratio',            type=float,
                    default=3,                    help='noise of input signal')

parser.add_argument('--NOISE_LEVEL',               type=float,
                    default=0.05,                    help='noise of input signal')
parser.add_argument('--WARP_FACTOR',               type=float,
                    default=0.1,                    help='noise of input signal')
parser.add_argument('--SFR_down',               type=float,
                    default=0.9,                    help='noise of input signal')
parser.add_argument('--SFR_up',               type=float,
                    default=1.1,                    help='noise of input signal')
parser.add_argument('--CROP_SIZE',               type=int,
                    default=50,                   help='noise of input signal')

parser.add_argument('--IN_test',               type=float,
                    default=0.,                    help='noise of input signal for test')
parser.add_argument('--R_train',               type=int,       default=1,
                    help='number of sampling for input noise in training')
parser.add_argument('--R_test',                type=int,       default=1,
                    help='number of sampling for input noise in testing')
# temporal information processing
parser.add_argument('--N_time',                type=int,       default=32,
                    help='number of sampling for temporal information processing')
parser.add_argument('--coupling_factor',       type=float,     default=0.3,
                    help='noise that caused by the circuit coupling')
# regularization
parser.add_argument('--pathnorm',              type=bool,      default=False,
                    help='path-norm as regularization for improving robustness against input noise')
# network-related
parser.add_argument('--hidden',                type=list,
                    default=[5],                   help='topology of the hidden layers')
# training-related
parser.add_argument('--SEED',                  type=int,
                    default=0,                     help='random seed')
parser.add_argument('--DEVICE',                type=str,
                    default='gpu',                 help='device for training')
parser.add_argument('--PATIENCE',              type=int,
                    default=500,                   help='patience for early-stopping')
parser.add_argument('--EPOCH',                 type=int,
                    default=10**10,                help='maximal epochs')
parser.add_argument('--LR',                    type=float,
                    default=0.1,                   help='learning rate')
parser.add_argument('--opt',                    type=str,
                    default='adam',                   help='type of optimizer')
parser.add_argument('--PROGRESSIVE',           type=bool,      default=True,
                    help='whether the learning rate will be adjusted')
parser.add_argument('--LR_PATIENCE',           type=int,       default=100,
                    help='patience for updating learning rate')
parser.add_argument('--LR_DECAY',              type=float,     default=0.5,
                    help='decay of learning rate for progressive lr')
parser.add_argument('--LR_MIN',                type=float,     default=1e-4,
                    help='minimal learning rate for stop training')
# metrics
parser.add_argument('--metric',                type=str,       default='acc',
                    help='nominal accuracy or measuring-aware accuracy')
parser.add_argument('--loss',                  type=str,
                    default='celoss',              help='loss function')
# server-related
parser.add_argument('--TIMELIMITATION',        type=float,
                    default=45,                    help='maximal running time (in hour)')

# hardware-related hyperparameters
# aging-related hyperparameters
parser.add_argument('--MODE',                  type=str,
                    default='nominal',             help='training mode: aging, nominal')
parser.add_argument('--M_train',               type=int,       default=1,
                    help='number of stochastic aging models during training')
parser.add_argument('--K_train',               type=int,       default=1,
                    help='number of temporal sampling during training')
parser.add_argument('--M_test',                type=int,       default=1,
                    help='number of stochastic aging models for testing')
parser.add_argument('--K_test',                type=int,       default=1,
                    help='number of temporal sampling for testing')
parser.add_argument('--t_test_max',            type=int,
                    default=1,                     help='test time interval')
parser.add_argument('--integration',           type=str,       default='MC',
                    help='method for integration: Monte-Carlo, Gaussian Quadrature')
# variation-related hyperparameters
parser.add_argument('--N_train',               type=int,       default=10,
                    help='number of sampling for variation during training')
parser.add_argument('--e_train',               type=float,
                    default=0.1,                    help='variation during training')
parser.add_argument('--N_test',                type=int,       default=1,
                    help='number of sampling for variation for testing')
parser.add_argument('--e_test',                type=int,
                    default=0.,                    help='variation for testing')
# power
parser.add_argument('--powerestimator',        type=str,       default='none',
                    help='the penalty term for encouraging lower energy')
parser.add_argument('--powerbalance',          type=float,     default=0.,
                    help='the scaling term for energy vs. accuracy')
parser.add_argument('--estimatorbalance',      type=float,     default=0.,
                    help='the scaling term for energy & weight decay')
parser.add_argument('--pgmin',                 type=float,     default=1e-7,
                    help='minimal printable conductance gmin')
# circuit learnability
parser.add_argument('--lnc',                   type=bool,      default=False,
                    help='shared learnable nonlinear components')
parser.add_argument('--ilnc',                  type=bool,      default=False,
                    help='independent learnable nonlinear components')


# log-file-related information
parser.add_argument('--projectname',           type=str,
                    default='project',             help='name of the project')
parser.add_argument('--temppath',              type=str,
                    default='/temp',               help='path to temp files')
parser.add_argument('--logfilepath',           type=str,
                    default='/log',                help='path to log files')
parser.add_argument('--report_freq',           type=int,
                    default=1,                     help='report frequency')
parser.add_argument('--recording',             type=bool,
                    default=False,                 help='save information in each epoch')
parser.add_argument('--recordpath',            type=str,
                    default='/record',             help='save information in each epoch')
parser.add_argument('--savepath',              type=str,
                    default='/experiment',         help='save information in each epoch')
parser.add_argument('--loglevel',              type=str,
                    default='info',                help='level of message logger')