junshan-kit 2.3.8__py2.py3-none-any.whl → 2.4.0__py2.py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- junshan_kit/DataSets.py +1 -1
- junshan_kit/ExperimentHub.py +194 -0
- junshan_kit/Models.py +207 -0
- junshan_kit/Optimizers.py +85 -0
- junshan_kit/Print_Info.py +59 -0
- junshan_kit/TrainingParas.py +470 -0
- junshan_kit/check_args.py +116 -0
- junshan_kit/datahub.py +281 -0
- junshan_kit/kit.py +7 -1
- {junshan_kit-2.3.8.dist-info → junshan_kit-2.4.0.dist-info}/METADATA +1 -3
- junshan_kit-2.4.0.dist-info/RECORD +14 -0
- junshan_kit-2.3.8.dist-info/RECORD +0 -7
- {junshan_kit-2.3.8.dist-info → junshan_kit-2.4.0.dist-info}/WHEEL +0 -0
|
@@ -0,0 +1,470 @@
|
|
|
1
|
+
# Step 1 : training_group
|
|
2
|
+
def set_training_group():
|
|
3
|
+
# <training_group>
|
|
4
|
+
training_group = [
|
|
5
|
+
# *********************************************************
|
|
6
|
+
# ----------------- MNIST (ResNet18) ----------------------
|
|
7
|
+
# ("ResNet18", "MNIST", "SGD"),
|
|
8
|
+
# ("ResNet18", "MNIST", "ADAM"),
|
|
9
|
+
# ("ResNet18", "MNIST", "SPSmax"),
|
|
10
|
+
# ("ResNet18", "MNIST", "Bundle"),
|
|
11
|
+
# ("ResNet18", "MNIST", "ALR-SMAG"),
|
|
12
|
+
# ("ResNet18", "MNIST", "SPBM-TR"),
|
|
13
|
+
# ("ResNet18", "MNIST", "SPBM-PF"),
|
|
14
|
+
# ("ResNet18", "MNIST", "SPBM-PF-NoneLower"),
|
|
15
|
+
# ("ResNet18", "MNIST", "SPBM-TR-NoneLower"),
|
|
16
|
+
# ("ResNet18", "MNIST", "SPBM-TR-NoneSpecial"),
|
|
17
|
+
# ---------------- CIFAR100 (ResNet18)---------------------
|
|
18
|
+
# ("ResNet18", "CIFAR100", "SGD"),
|
|
19
|
+
# ("ResNet18", "CIFAR100", "ADAM"),
|
|
20
|
+
# ("ResNet18", "CIFAR100", "SPSmax"),
|
|
21
|
+
# ("ResNet18", "CIFAR100", "Bundle"),
|
|
22
|
+
# ("ResNet18", "CIFAR100", "ALR-SMAG"),
|
|
23
|
+
# ("ResNet18", "CIFAR100", "SPBM-TR"),
|
|
24
|
+
# ("ResNet18", "CIFAR100", "SPBM-PF"),
|
|
25
|
+
# ("ResNet18", "CIFAR100", "SPBM-PF-NoneLower"),
|
|
26
|
+
# ("ResNet18", "CIFAR100", "SPBM-TR-NoneLower"),
|
|
27
|
+
# # ----------- CALTECH101_Resize_32 (ResNet18) -------------
|
|
28
|
+
("ResNet18", "CALTECH101_Resize_32", "SGD"),
|
|
29
|
+
# ("ResNet18", "CALTECH101_Resize_32", "ADAM"),
|
|
30
|
+
# ("ResNet18", "CALTECH101_Resize_32", "SPSmax"),
|
|
31
|
+
# ("ResNet18", "CALTECH101_Resize_32", "Bundle"),
|
|
32
|
+
# ("ResNet18", "CALTECH101_Resize_32", "ALR-SMAG"),
|
|
33
|
+
# ("ResNet18", "CALTECH101_Resize_32", "SPBM-TR"),
|
|
34
|
+
# ("ResNet18", "CALTECH101_Resize_32", "SPBM-PF"),
|
|
35
|
+
# ("ResNet18", "CALTECH101_Resize_32", "SPBM-PF-NoneLower"),
|
|
36
|
+
# ("ResNet18", "CALTECH101_Resize_32", "SPBM-TR-NoneLower"),
|
|
37
|
+
|
|
38
|
+
# *********************************************************
|
|
39
|
+
# ---------------- MNIST (ResNet34) -----------------------
|
|
40
|
+
# ("ResNet34" ,"MNIST", "SGD"),
|
|
41
|
+
# ("ResNet34" ,"MNIST", "ADAM"),
|
|
42
|
+
# ("ResNet34" ,"MNIST", "SPSmax"),
|
|
43
|
+
# ("ResNet34" ,"MNIST", "Bundle"),
|
|
44
|
+
# ("ResNet34" ,"MNIST", "ALR-SMAG"),
|
|
45
|
+
# ("ResNet34" ,"MNIST", "SPBM-TR"),
|
|
46
|
+
# ("ResNet34" ,"MNIST", "SPBM-PF"),
|
|
47
|
+
# ("ResNet34" ,"MNIST", "SPBM-PF-NoneLower"),
|
|
48
|
+
# ("ResNet34" ,"MNIST", "SPBM-TR-NoneLower"),
|
|
49
|
+
# ------------------ CIFAR100 (ResNet34)-------------------
|
|
50
|
+
# ("ResNet34" ,"CIFAR100", "SGD"),
|
|
51
|
+
# ("ResNet34" ,"CIFAR100", "ADAM"),
|
|
52
|
+
# ("ResNet34" ,"CIFAR100", "SPSmax"),
|
|
53
|
+
# ("ResNet34" ,"CIFAR100", "Bundle"),
|
|
54
|
+
# ("ResNet34" ,"CIFAR100", "ALR-SMAG"),
|
|
55
|
+
# ("ResNet34" ,"CIFAR100", "SPBM-TR"),
|
|
56
|
+
# ("ResNet34" ,"CIFAR100", "SPBM-PF"),
|
|
57
|
+
# ("ResNet34" ,"CIFAR100", "SPBM-PF-NoneLower"),
|
|
58
|
+
# ("ResNet34" ,"CIFAR100", "SPBM-TR-NoneLower"),
|
|
59
|
+
# ------------ CALTECH101_Resize_32 (ResNet34) ------------
|
|
60
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "SGD"),
|
|
61
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "ADAM"),
|
|
62
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "SPSmax"),
|
|
63
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "Bundle"),
|
|
64
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "ALR-SMAG"),
|
|
65
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "SPBM-TR"),
|
|
66
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "SPBM-PF"),
|
|
67
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "SPBM-PF-NoneLower"),
|
|
68
|
+
# ("ResNet34" ,"CALTECH101_Resize_32", "SPBM-TR-NoneLower"),
|
|
69
|
+
|
|
70
|
+
# *********************************************************
|
|
71
|
+
# ------------------ MNIST (LeastSquares) -----------------
|
|
72
|
+
# ("LeastSquares" ,"MNIST", "SGD"),
|
|
73
|
+
# ("LeastSquares" ,"MNIST", "ADAM"),
|
|
74
|
+
# ("LeastSquares" ,"MNIST", "SPSmax"),
|
|
75
|
+
# ("LeastSquares" ,"MNIST", "Bundle"),
|
|
76
|
+
# ("LeastSquares" ,"MNIST", "ALR-SMAG"),
|
|
77
|
+
# ("LeastSquares" ,"MNIST", "SPBM-TR"),
|
|
78
|
+
# ("LeastSquares" ,"MNIST", "SPBM-PF"),
|
|
79
|
+
# ("LeastSquares" ,"MNIST", "SPBM-PF-NoneLower"),
|
|
80
|
+
# ("LeastSquares" ,"MNIST", "SPBM-TR-NoneLower"),
|
|
81
|
+
# ---------------- CIFAR100 (LeastSquares) ----------------
|
|
82
|
+
# ("LeastSquares" ,"CIFAR100", "SGD"),
|
|
83
|
+
# ("LeastSquares" ,"CIFAR100", "ADAM"),
|
|
84
|
+
# ("LeastSquares" ,"CIFAR100", "SPSmax"),
|
|
85
|
+
# ("LeastSquares" ,"CIFAR100", "Bundle"),
|
|
86
|
+
# ("LeastSquares" ,"CIFAR100", "ALR-SMAG"),
|
|
87
|
+
# ("LeastSquares" ,"CIFAR100", "SPBM-TR"),
|
|
88
|
+
# ("LeastSquares" ,"CIFAR100", "SPBM-PF"),
|
|
89
|
+
# ("LeastSquares" ,"CIFAR100", "SPBM-PF-NoneLower"),
|
|
90
|
+
# ("LeastSquares" ,"CIFAR100", "SPBM-TR-NoneLower"),
|
|
91
|
+
# ---------------- CIFAR100 (LeastSquares) ----------------
|
|
92
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "SGD"),
|
|
93
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "ADAM"),
|
|
94
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "SPSmax"),
|
|
95
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "Bundle"),
|
|
96
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "ALR-SMAG"),
|
|
97
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "SPBM-TR"),
|
|
98
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "SPBM-PF"),
|
|
99
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "SPBM-PF-NoneLower"),
|
|
100
|
+
# ("LeastSquares" ,"CALTECH101_Resize_32", "SPBM-TR-NoneLower"),
|
|
101
|
+
|
|
102
|
+
# *********************************************************
|
|
103
|
+
# ------------- MNIST (LogRegressionBinary) ---------------
|
|
104
|
+
# ("LogRegressionBinary" ,"MNIST", "SGD"),
|
|
105
|
+
# ("LogRegressionBinary" ,"MNIST", "ADAM"),
|
|
106
|
+
# ("LogRegressionBinary" ,"MNIST", "SPSmax"),
|
|
107
|
+
# ("LogRegressionBinary" ,"MNIST", "Bundle"),
|
|
108
|
+
# ("LogRegressionBinary" ,"MNIST", "ALR-SMAG"),
|
|
109
|
+
# ("LogRegressionBinary" ,"MNIST", "SPBM-TR"),
|
|
110
|
+
# ("LogRegressionBinary" ,"MNIST", "SPBM-PF"),
|
|
111
|
+
# ("LogRegressionBinary" ,"MNIST", "SPBM-PF-NoneLower"),
|
|
112
|
+
# ("LogRegressionBinary" ,"MNIST", "SPBM-TR-NoneLower"),
|
|
113
|
+
# ------------- CIFAR100 (LogRegressionBinary) ------------
|
|
114
|
+
# ("LogRegressionBinary" ,"CIFAR100", "SGD"),
|
|
115
|
+
# ("LogRegressionBinary" ,"CIFAR100", "ADAM"),
|
|
116
|
+
# ("LogRegressionBinary" ,"CIFAR100", "SPSmax"),
|
|
117
|
+
# ("LogRegressionBinary" ,"CIFAR100", "Bundle"),
|
|
118
|
+
# ("LogRegressionBinary" ,"CIFAR100", "ALR-SMAG"),
|
|
119
|
+
# ("LogRegressionBinary" ,"CIFAR100", "SPBM-TR"),
|
|
120
|
+
# ("LogRegressionBinary" ,"CIFAR100", "SPBM-PF"),
|
|
121
|
+
# ("LogRegressionBinary" ,"CIFAR100", "SPBM-PF-NoneLower"),
|
|
122
|
+
# ("LogRegressionBinary" ,"CIFAR100", "SPBM-TR-NoneLower"),
|
|
123
|
+
# # --------------- RCV1 (LogRegressionBinary) --------------
|
|
124
|
+
# ("LogRegressionBinary" ,"RCV1", "SGD"),
|
|
125
|
+
# ("LogRegressionBinary" ,"RCV1", "ADAM"),
|
|
126
|
+
# ("LogRegressionBinary" ,"RCV1", "SPSmax"),
|
|
127
|
+
# ("LogRegressionBinary" ,"RCV1", "Bundle"),
|
|
128
|
+
# ("LogRegressionBinary" ,"RCV1", "ALR-SMAG"),
|
|
129
|
+
# ("LogRegressionBinary" ,"RCV1", "SPBM-TR"),
|
|
130
|
+
# ("LogRegressionBinary" ,"RCV1", "SPBM-PF"),
|
|
131
|
+
|
|
132
|
+
# # *********************************************************
|
|
133
|
+
# # ------------ MNIST (LogRegressionBinaryL2) --------------
|
|
134
|
+
# ("LogRegressionBinaryL2" ,"MNIST", "SGD"),
|
|
135
|
+
# ("LogRegressionBinaryL2" ,"MNIST", "ADAM"),
|
|
136
|
+
# ("LogRegressionBinaryL2" ,"MNIST", "SPSmax"),
|
|
137
|
+
# ("LogRegressionBinaryL2" ,"MNIST", "Bundle"),
|
|
138
|
+
# ("LogRegressionBinaryL2" ,"MNIST", "ALR-SMAG"),
|
|
139
|
+
# ("LogRegressionBinaryL2" ,"MNIST", "SPBM-TR"),
|
|
140
|
+
# ("LogRegressionBinaryL2" ,"MNIST", "SPBM-PF"),
|
|
141
|
+
# # ------------- CIFAR100 (LogRegressionBinaryL2) ----------
|
|
142
|
+
# ("LogRegressionBinaryL2" ,"CIFAR100", "SGD"),
|
|
143
|
+
# ("LogRegressionBinaryL2" ,"CIFAR100", "ADAM"),
|
|
144
|
+
# ("LogRegressionBinaryL2" ,"CIFAR100", "SPSmax"),
|
|
145
|
+
# ("LogRegressionBinaryL2" ,"CIFAR100", "Bundle"),
|
|
146
|
+
# ("LogRegressionBinaryL2" ,"CIFAR100", "ALR-SMAG"),
|
|
147
|
+
# ("LogRegressionBinaryL2" ,"CIFAR100", "SPBM-TR"),
|
|
148
|
+
# ("LogRegressionBinaryL2" ,"CIFAR100", "SPBM-PF"),
|
|
149
|
+
# # --------------- RCV1 (LogRegressionBinaryL2) ------------
|
|
150
|
+
# ("LogRegressionBinaryL2", "RCV1", "SGD"),
|
|
151
|
+
# ("LogRegressionBinaryL2", "RCV1", "ADAM"),
|
|
152
|
+
# ("LogRegressionBinaryL2", "RCV1", "SPSmax"),
|
|
153
|
+
# ("LogRegressionBinaryL2", "RCV1", "Bundle"),
|
|
154
|
+
# ("LogRegressionBinaryL2", "RCV1", "ALR-SMAG"),
|
|
155
|
+
# ("LogRegressionBinaryL2", "RCV1", "SPBM-TR"),
|
|
156
|
+
# ("LogRegressionBinaryL2", "RCV1", "SPBM-PF"),
|
|
157
|
+
# # -------------- Duke (LogRegressionBinaryL2) -------------
|
|
158
|
+
# ("LogRegressionBinaryL2" ,"Duke", "SGD"),
|
|
159
|
+
# ("LogRegressionBinaryL2" ,"Duke", "ADAM"),
|
|
160
|
+
# ("LogRegressionBinaryL2" ,"Duke", "SPSmax"),
|
|
161
|
+
# ("LogRegressionBinaryL2" ,"Duke", "Bundle"),
|
|
162
|
+
# ("LogRegressionBinaryL2" ,"Duke", "ALR-SMAG"),
|
|
163
|
+
# ("LogRegressionBinaryL2" ,"Duke", "SPBM-TR"),
|
|
164
|
+
# ("LogRegressionBinaryL2" ,"Duke", "SPBM-PF"),
|
|
165
|
+
# # -------------- Ijcnn (LogRegressionBinaryL2) ------------
|
|
166
|
+
# ("LogRegressionBinaryL2", "Ijcnn", "SGD"),
|
|
167
|
+
# ("LogRegressionBinaryL2", "Ijcnn", "ADAM"),
|
|
168
|
+
# ("LogRegressionBinaryL2", "Ijcnn", "SPSmax"),
|
|
169
|
+
# ("LogRegressionBinaryL2", "Ijcnn", "Bundle"),
|
|
170
|
+
# ("LogRegressionBinaryL2", "Ijcnn", "ALR-SMAG"),
|
|
171
|
+
# ("LogRegressionBinaryL2", "Ijcnn", "SPBM-TR"),
|
|
172
|
+
# ("LogRegressionBinaryL2", "Ijcnn", "SPBM-PF"),
|
|
173
|
+
# # ----------------- w8a (LogRegressionBinaryL2) -----------
|
|
174
|
+
# ("LogRegressionBinaryL2", "w8a", "SGD"),
|
|
175
|
+
# ("LogRegressionBinaryL2", "w8a", "ADAM"),
|
|
176
|
+
# ("LogRegressionBinaryL2", "w8a", "SPSmax"),
|
|
177
|
+
# ("LogRegressionBinaryL2", "w8a", "Bundle"),
|
|
178
|
+
# ("LogRegressionBinaryL2", "w8a", "ALR-SMAG"),
|
|
179
|
+
# ("LogRegressionBinaryL2", "w8a", "SPBM-TR"),
|
|
180
|
+
# ("LogRegressionBinaryL2", "w8a", "SPBM-PF"),
|
|
181
|
+
]
|
|
182
|
+
# <training_group>
|
|
183
|
+
|
|
184
|
+
return training_group
|
|
185
|
+
|
|
186
|
+
def batch_size() -> dict:
|
|
187
|
+
batch_size = {
|
|
188
|
+
# 15123/12560
|
|
189
|
+
"Shuttle": 256,
|
|
190
|
+
# 15000/5000
|
|
191
|
+
"Letter": 256,
|
|
192
|
+
# 528/462
|
|
193
|
+
"Vowel": 52,
|
|
194
|
+
# 60000/10000
|
|
195
|
+
"MNIST": 256,
|
|
196
|
+
# 50000/10000
|
|
197
|
+
"CIFAR100": 256,
|
|
198
|
+
# 8,677 (will be split into 7:3)---> 6073/2604
|
|
199
|
+
"CALTECH101_Resize_32": 256,
|
|
200
|
+
# 20,242 (will be split into 7:3)---> 14169/6073
|
|
201
|
+
"RCV1": 256,
|
|
202
|
+
# only 42 (38+4) examples (cancer data)
|
|
203
|
+
"Duke": 10,
|
|
204
|
+
# 35000 + 91701
|
|
205
|
+
"Ijcnn": 64,
|
|
206
|
+
# classes: 2 data: (49749 14,951) features: 300
|
|
207
|
+
"w8a": 128,
|
|
208
|
+
}
|
|
209
|
+
return batch_size
|
|
210
|
+
|
|
211
|
+
|
|
212
|
+
def epochs(OtherParas) -> dict:
|
|
213
|
+
epochs = {
|
|
214
|
+
# 15123/12560
|
|
215
|
+
"Shuttle": 10,
|
|
216
|
+
# 15000/5000
|
|
217
|
+
"Letter": 10,
|
|
218
|
+
# 528/462
|
|
219
|
+
"Vowel": 10,
|
|
220
|
+
# 60000/10000
|
|
221
|
+
"MNIST": 50,
|
|
222
|
+
# 50000/10000
|
|
223
|
+
"CIFAR100": 50,
|
|
224
|
+
# 8,677 (will be split into 7:3)---> 6073/2604
|
|
225
|
+
"CALTECH101_Resize_32": 50,
|
|
226
|
+
# 20,242 (will be split into 7:3)---> 14169/6073
|
|
227
|
+
"RCV1": 10,
|
|
228
|
+
# only 42 (38+4) examples (cancer data)
|
|
229
|
+
"Duke": 10,
|
|
230
|
+
# 35000 + 91701
|
|
231
|
+
"Ijcnn": 10,
|
|
232
|
+
# classes: 2 data: (49749 14,951) features: 300
|
|
233
|
+
"w8a": 10,
|
|
234
|
+
}
|
|
235
|
+
if OtherParas["debug"]:
|
|
236
|
+
epochs = {k: 2 for k in epochs}
|
|
237
|
+
|
|
238
|
+
return epochs
|
|
239
|
+
|
|
240
|
+
|
|
241
|
+
def split_train_data() -> dict:
|
|
242
|
+
split_train_data = {
|
|
243
|
+
# 20,242 + 0 (test data to large)
|
|
244
|
+
"RCV1": 0.7,
|
|
245
|
+
# only 42 (38+4) examples (Not need)
|
|
246
|
+
"Duke": 1,
|
|
247
|
+
# classes: 2 data: (35000, 91701) features: 22
|
|
248
|
+
"Ijcnn": 1,
|
|
249
|
+
# classes: 2 data: (49749 14,951) features: 300
|
|
250
|
+
"w8a": 1,
|
|
251
|
+
}
|
|
252
|
+
return split_train_data
|
|
253
|
+
|
|
254
|
+
|
|
255
|
+
def select_subset():
|
|
256
|
+
select_subset = {
|
|
257
|
+
"CALTECH101_Resize_32": True,
|
|
258
|
+
"CIFAR100": True,
|
|
259
|
+
"Duke": False,
|
|
260
|
+
"Ijcnn": True,
|
|
261
|
+
"MNIST": True,
|
|
262
|
+
"RCV1": True,
|
|
263
|
+
"w8a": True,
|
|
264
|
+
}
|
|
265
|
+
return select_subset
|
|
266
|
+
|
|
267
|
+
|
|
268
|
+
def subset_number_dict(OtherParas):
|
|
269
|
+
subset_number_dict = {
|
|
270
|
+
# Max: 60,000/10,000
|
|
271
|
+
"MNIST": (1000, 5000),
|
|
272
|
+
# Max: 50,000
|
|
273
|
+
"CIFAR100": (2000, 10000),
|
|
274
|
+
# Max: 8,677 (6073/2604)
|
|
275
|
+
"CALTECH101_Resize_32": (2000, 2604), #test max: 2604
|
|
276
|
+
# classes: 2 data: (35000, 91701) features: 22
|
|
277
|
+
"Ijcnn": (1000, 1000),
|
|
278
|
+
# classes: 2 data: (14169, 6,073) features: 47,236
|
|
279
|
+
"RCV1": (1000, 1000),
|
|
280
|
+
# classes: 2 data: (49749 14,951) features: 300
|
|
281
|
+
"w8a": (1000, 1000),
|
|
282
|
+
}
|
|
283
|
+
|
|
284
|
+
if OtherParas["debug"]:
|
|
285
|
+
subset_number_dict = {k: (50, 50) for k in subset_number_dict}
|
|
286
|
+
return subset_number_dict
|
|
287
|
+
|
|
288
|
+
|
|
289
|
+
def validation() -> dict:
|
|
290
|
+
validation = {
|
|
291
|
+
# "MNIST": True,
|
|
292
|
+
# "CIFAR100": True,
|
|
293
|
+
# "CALTECH101_Resize_32": True
|
|
294
|
+
}
|
|
295
|
+
return validation
|
|
296
|
+
|
|
297
|
+
|
|
298
|
+
def validation_rate() -> dict:
|
|
299
|
+
validation_rate = {
|
|
300
|
+
"MNIST": 0.3, # Max: 60,000/10,000
|
|
301
|
+
"CIFAR100": 0.3, # Max: 50,000
|
|
302
|
+
"CALTECH101_Resize_32": 0.3, # Max: 8,677 (6073/2604)
|
|
303
|
+
}
|
|
304
|
+
return validation_rate
|
|
305
|
+
|
|
306
|
+
|
|
307
|
+
def model_list() -> list:
|
|
308
|
+
model_list = [
|
|
309
|
+
"ResNet18",
|
|
310
|
+
"ResNet34",
|
|
311
|
+
"LeastSquares",
|
|
312
|
+
"LogRegressionBinary",
|
|
313
|
+
"LogRegressionBinaryL2",
|
|
314
|
+
]
|
|
315
|
+
return model_list
|
|
316
|
+
|
|
317
|
+
|
|
318
|
+
def model_type() -> dict:
|
|
319
|
+
model_type = {
|
|
320
|
+
"ResNet18": "multi",
|
|
321
|
+
"ResNet34": "multi",
|
|
322
|
+
"LeastSquares": "multi",
|
|
323
|
+
"LogRegressionBinary": "binary",
|
|
324
|
+
"LogRegressionBinaryL2": "binary",
|
|
325
|
+
}
|
|
326
|
+
return model_type
|
|
327
|
+
|
|
328
|
+
|
|
329
|
+
def data_list() -> list:
|
|
330
|
+
data_list = [
|
|
331
|
+
# classes: 2 data: 42 (38+4) features: 7,129
|
|
332
|
+
"Duke",
|
|
333
|
+
# classes: 2 data: (35000, 91701) features: 22
|
|
334
|
+
"Ijcnn",
|
|
335
|
+
# classes: 2 data: (49749 14,951) features: 300
|
|
336
|
+
"w8a",
|
|
337
|
+
#
|
|
338
|
+
"RCV1",
|
|
339
|
+
"Shuttle",
|
|
340
|
+
"Letter",
|
|
341
|
+
"Vowel",
|
|
342
|
+
"MNIST",
|
|
343
|
+
"CIFAR100",
|
|
344
|
+
"CALTECH101_Resize_32",
|
|
345
|
+
]
|
|
346
|
+
return data_list
|
|
347
|
+
|
|
348
|
+
def optimizer_dict(OtherParas)->dict:
|
|
349
|
+
optimizer_dict = {
|
|
350
|
+
# --------------------------- ADAM ----------------------------
|
|
351
|
+
"ADAM": {
|
|
352
|
+
"params": {
|
|
353
|
+
# "alpha": [2 * 1e-3],
|
|
354
|
+
"alpha": (
|
|
355
|
+
[0.5 * 1e-3, 1e-3, 2 * 1e-3]
|
|
356
|
+
if OtherParas["SeleParasOn"]
|
|
357
|
+
else [0.0005]
|
|
358
|
+
),
|
|
359
|
+
"epsilon": [1e-8],
|
|
360
|
+
"beta1": [0.9],
|
|
361
|
+
"beta2": [0.999],
|
|
362
|
+
},
|
|
363
|
+
},
|
|
364
|
+
# ----------------------- ALR-SMAG ---------------------------
|
|
365
|
+
"ALR-SMAG": {
|
|
366
|
+
"params": {
|
|
367
|
+
"c": ([0.1, 0.5, 1, 5, 10] if OtherParas["SeleParasOn"] else [0.1]),
|
|
368
|
+
"eta_max": (
|
|
369
|
+
[2**i for i in range(-8, 9)]
|
|
370
|
+
if OtherParas["SeleParasOn"]
|
|
371
|
+
else [0.125]
|
|
372
|
+
),
|
|
373
|
+
"beta": [0.9],
|
|
374
|
+
},
|
|
375
|
+
},
|
|
376
|
+
# ------------------------ Bundle -----------------------------
|
|
377
|
+
"Bundle": {
|
|
378
|
+
"params": {
|
|
379
|
+
"delta": (
|
|
380
|
+
[2**i for i in range(-8, 9)]
|
|
381
|
+
if OtherParas["SeleParasOn"]
|
|
382
|
+
else [0.25]
|
|
383
|
+
),
|
|
384
|
+
"cutting_number": [10],
|
|
385
|
+
},
|
|
386
|
+
},
|
|
387
|
+
# --------------------------- SGD -----------------------------
|
|
388
|
+
"SGD": {
|
|
389
|
+
"params": {
|
|
390
|
+
"alpha": (
|
|
391
|
+
[2**i for i in range(-8, 9)] if OtherParas["SeleParasOn"] else [0.1]
|
|
392
|
+
)
|
|
393
|
+
}
|
|
394
|
+
},
|
|
395
|
+
# -------------------------- SPSmax ---------------------------
|
|
396
|
+
"SPSmax": {
|
|
397
|
+
"params": {
|
|
398
|
+
"c": ([0.1, 0.5, 1, 5, 10] if OtherParas["SeleParasOn"] else [0.1]),
|
|
399
|
+
"gamma": (
|
|
400
|
+
[2**i for i in range(-8, 9)]
|
|
401
|
+
if OtherParas["SeleParasOn"]
|
|
402
|
+
else [0.125]),
|
|
403
|
+
},
|
|
404
|
+
},
|
|
405
|
+
# ----------------------- SPBM-PF -----------------------------
|
|
406
|
+
"SPBM-PF": {
|
|
407
|
+
"params": {
|
|
408
|
+
"M": [1e-5],
|
|
409
|
+
"delta": (
|
|
410
|
+
[2**i for i in range(9, 20)]
|
|
411
|
+
if OtherParas["SeleParasOn"]
|
|
412
|
+
else [1]
|
|
413
|
+
),
|
|
414
|
+
"cutting_number": [10],
|
|
415
|
+
},
|
|
416
|
+
},
|
|
417
|
+
# ----------------------- SPBM-TR -----------------------------
|
|
418
|
+
"SPBM-TR": {
|
|
419
|
+
"params": {
|
|
420
|
+
"M": [1e-5],
|
|
421
|
+
"delta": (
|
|
422
|
+
[2**i for i in range(9, 20)]
|
|
423
|
+
if OtherParas["SeleParasOn"]
|
|
424
|
+
else [256]
|
|
425
|
+
),
|
|
426
|
+
"cutting_number": [10],
|
|
427
|
+
},
|
|
428
|
+
},
|
|
429
|
+
|
|
430
|
+
# ------------------- SPBM-TR-NoneLower -----------------------
|
|
431
|
+
"SPBM-TR-NoneLower": {
|
|
432
|
+
"params": {
|
|
433
|
+
"M": [1e-5],
|
|
434
|
+
"delta": (
|
|
435
|
+
[2**i for i in range(0, 9)]
|
|
436
|
+
if OtherParas["SeleParasOn"]
|
|
437
|
+
else [256]
|
|
438
|
+
),
|
|
439
|
+
"cutting_number": [10],
|
|
440
|
+
},
|
|
441
|
+
},
|
|
442
|
+
# ------------------- SPBM-TR-NoneSpecial -----------------------
|
|
443
|
+
"SPBM-TR-NoneSpecial": {
|
|
444
|
+
"params": {
|
|
445
|
+
"M": [1e-5],
|
|
446
|
+
"delta": (
|
|
447
|
+
[2**i for i in range(-8, 9)]
|
|
448
|
+
if OtherParas["SeleParasOn"]
|
|
449
|
+
else [1]
|
|
450
|
+
),
|
|
451
|
+
"cutting_number": [10],
|
|
452
|
+
},
|
|
453
|
+
},
|
|
454
|
+
# -------------------- SPBM-PF-NoneLower ----------------------
|
|
455
|
+
"SPBM-PF-NoneLower": {
|
|
456
|
+
"params": {
|
|
457
|
+
"M": [1e-5],
|
|
458
|
+
"delta": (
|
|
459
|
+
[2**i for i in range(0, 9)]
|
|
460
|
+
if OtherParas["SeleParasOn"]
|
|
461
|
+
else [0]
|
|
462
|
+
),
|
|
463
|
+
"cutting_number": [10],
|
|
464
|
+
},
|
|
465
|
+
},
|
|
466
|
+
|
|
467
|
+
|
|
468
|
+
}
|
|
469
|
+
return optimizer_dict
|
|
470
|
+
|
|
@@ -0,0 +1,116 @@
|
|
|
1
|
+
import argparse
|
|
2
|
+
from junshan_kit import Models
|
|
3
|
+
|
|
4
|
+
def get_args():
|
|
5
|
+
parser = argparse.ArgumentParser(description="Combined config argument example")
|
|
6
|
+
|
|
7
|
+
allowed_models = ["LS", "LRL2","ResNet18"]
|
|
8
|
+
allowed_optimizers = ["Adam", "SGD",]
|
|
9
|
+
allowed_datasets = ["MNIST", "CIFAR100"]
|
|
10
|
+
|
|
11
|
+
model_mapping = {
|
|
12
|
+
"LS": "LeastSquares",
|
|
13
|
+
"LRL2": "LogRegressionBinaryL2",
|
|
14
|
+
"ResNet18": "ResNet18"
|
|
15
|
+
}
|
|
16
|
+
|
|
17
|
+
# Single combined argument that can appear multiple times
|
|
18
|
+
parser.add_argument(
|
|
19
|
+
"--train_group",
|
|
20
|
+
type=str,
|
|
21
|
+
nargs="+", # Allow multiple configs
|
|
22
|
+
required=True,
|
|
23
|
+
help = f"Format: model-dataset-optimizer (e.g., ResNet18-CIFAR10-Adam). model: {model_mapping}, \n datasets: {allowed_datasets}, optimizers: {allowed_optimizers},"
|
|
24
|
+
)
|
|
25
|
+
|
|
26
|
+
parser.add_argument(
|
|
27
|
+
"--e",
|
|
28
|
+
type=int,
|
|
29
|
+
required=True,
|
|
30
|
+
help="Number of training epochs. Example: --e 50"
|
|
31
|
+
)
|
|
32
|
+
|
|
33
|
+
parser.add_argument(
|
|
34
|
+
"--seed",
|
|
35
|
+
type=int,
|
|
36
|
+
default=42,
|
|
37
|
+
help="Random seed for experiment reproducibility. Default: 42"
|
|
38
|
+
)
|
|
39
|
+
|
|
40
|
+
parser.add_argument(
|
|
41
|
+
"--bs",
|
|
42
|
+
type=int,
|
|
43
|
+
required=True,
|
|
44
|
+
help="Batch size for training. Example: --bs 128"
|
|
45
|
+
)
|
|
46
|
+
|
|
47
|
+
parser.add_argument(
|
|
48
|
+
"--cuda",
|
|
49
|
+
type=int,
|
|
50
|
+
default=0,
|
|
51
|
+
required=True,
|
|
52
|
+
help="The number of cuda. Example: --cuda 1 (default=0) "
|
|
53
|
+
)
|
|
54
|
+
|
|
55
|
+
parser.add_argument(
|
|
56
|
+
"--s",
|
|
57
|
+
type=float,
|
|
58
|
+
default=1.0,
|
|
59
|
+
# required=True,
|
|
60
|
+
help="Proportion of dataset to use for training split. Example: --s 0.8 (default=1.0)"
|
|
61
|
+
)
|
|
62
|
+
|
|
63
|
+
parser.add_argument(
|
|
64
|
+
"--subset",
|
|
65
|
+
type=float,
|
|
66
|
+
nargs=2,
|
|
67
|
+
# required=True,
|
|
68
|
+
help = "Two subset ratios (train, test), e.g., --subset 0.7 0.3 or --subset 500 500"
|
|
69
|
+
)
|
|
70
|
+
|
|
71
|
+
args = parser.parse_args()
|
|
72
|
+
args.model_mapping = model_mapping
|
|
73
|
+
|
|
74
|
+
|
|
75
|
+
if args.subset is not None:
|
|
76
|
+
check_subset_info(args, parser)
|
|
77
|
+
|
|
78
|
+
|
|
79
|
+
check_args(args, parser, allowed_models, allowed_optimizers, allowed_datasets, model_mapping)
|
|
80
|
+
|
|
81
|
+
return args
|
|
82
|
+
|
|
83
|
+
def check_subset_info(args, parser):
|
|
84
|
+
total = sum(args.subset)
|
|
85
|
+
if args.subset[0]>1:
|
|
86
|
+
# CHECK
|
|
87
|
+
for i in args.subset:
|
|
88
|
+
if i < 1:
|
|
89
|
+
parser.error(f"Invalid --subset {args.subset}: The number of subdata must > 1")
|
|
90
|
+
else:
|
|
91
|
+
if abs(total - 1.0) != 0.0:
|
|
92
|
+
parser.error(f"Invalid --subset {args.subset}: the values must sum to 1.0 (current sum = {total:.6f})")
|
|
93
|
+
|
|
94
|
+
|
|
95
|
+
def check_args(args, parser, allowed_models, allowed_optimizers, allowed_datasets, model_mapping):
|
|
96
|
+
# Parse and validate each train_group
|
|
97
|
+
for cfg in args.train_group:
|
|
98
|
+
try:
|
|
99
|
+
model, dataset, optimizer = cfg.split("-")
|
|
100
|
+
|
|
101
|
+
if model not in allowed_models:
|
|
102
|
+
parser.error(f"Invalid model '{model}'. Choose from {allowed_models}")
|
|
103
|
+
if optimizer not in allowed_optimizers:
|
|
104
|
+
parser.error(f"Invalid optimizer '{optimizer}'. Choose from {allowed_optimizers}")
|
|
105
|
+
if dataset not in allowed_datasets:
|
|
106
|
+
parser.error(f"Invalid dataset '{dataset}'. Choose from {allowed_datasets}")
|
|
107
|
+
|
|
108
|
+
except ValueError:
|
|
109
|
+
parser.error(f"Invalid format '{cfg}'. Use model-dataset-optimizer")
|
|
110
|
+
|
|
111
|
+
for cfg in args.train_group:
|
|
112
|
+
model_name, dataset_name, optimizer_name = cfg.split("-")
|
|
113
|
+
try:
|
|
114
|
+
f = getattr(Models, f"Build_{model_mapping[model_name]}_{dataset_name}")
|
|
115
|
+
except:
|
|
116
|
+
print(getattr(Models, f"Build_{model_mapping[model_name]}_{dataset_name}"))
|