-
Notifications
You must be signed in to change notification settings - Fork 2
/
experiments.py
345 lines (247 loc) · 13.9 KB
/
experiments.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
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
import propinfer as pia
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from os.path import join
import sys
from sklearn.linear_model import LogisticRegression
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import accuracy_score, mean_squared_error
from tqdm.notebook import tqdm, trange
from copy import deepcopy
import torch
import torch.nn as nn
from constants import N_RUNS, N_TARGETS, N_SHADOWS, DEEPSETS_HYPERPARAMS, MLP_HYPERPARAMS
from generators import NN_Generator, Binary_NN_Generator, Membership_Generator
from models import MLPEarlyStopping, IRM
def experiment_a(results_folder):
res = pd.DataFrame(columns=['epsilon', 'MAE/Acc', 'Experiment', 'Attack'])
perf = pd.DataFrame(columns=['epsilon', 'MSE', 'Experiment'])
# We keep a same neural network as a basis for D_0
base_nn_gen = NN_Generator(2**11)
# Sampling only D_0 training set, since the distribution of D_1 will change
df_0 = Binary_NN_Generator(base_nn_gen=base_nn_gen, epsilon=0.).sample(False)
for epsilon in np.arange(0., 0.21, 0.04):
print("Running experiment A with epsilon = {:.2f}".format(epsilon))
gen = Binary_NN_Generator(base_nn_gen=base_nn_gen, epsilon=epsilon)
# Experiment A - Supports classification only
exp_class = pia.Experiment(gen, 'label', pia.MLP, N_TARGETS, N_SHADOWS, MLP_HYPERPARAMS, n_classes=2)
exp_class.run_targets()
exp_class.run_shadows()
res.loc[len(res)] = (epsilon,
exp_class.run_whitebox_deepsets(DEEPSETS_HYPERPARAMS, N_RUNS),
'Classification',
'WhiteBox')
res.loc[len(res)] = (epsilon,
exp_class.run_blackbox(N_RUNS),
'Classification',
'BlackBox')
# Since E(Y|X) changes depending on the classification label, we only test the models against their own class test set
df_1 = gen.sample(True)
for t, target in enumerate(exp_class.targets):
if exp_class.labels[t]:
perf.loc[len(perf)] = (epsilon, mean_squared_error(df_1.label, target.predict(df_1)), 'Classification')
else:
perf.loc[len(perf)] = (epsilon, mean_squared_error(df_0.label, target.predict(df_0)), 'Classification')
res.explode('MAE/Acc').reset_index(drop=True).to_pickle(join(results_folder, 'exp_a_res.pkl'))
perf.to_pickle(join(results_folder, 'exp_a_perf.pkl'))
def experiment_abis(results_folder):
res = pd.DataFrame(columns=['early_stop', 'MAE/Acc', 'Experiment', 'Attack'])
perf = pd.DataFrame(columns=['early_stop', 'MSE', 'Experiment'])
base_nn_gen = NN_Generator(2**11)
gen = Binary_NN_Generator(base_nn_gen=base_nn_gen, epsilon=0.05)
# Sampling target models' test sets
df_0 = gen.sample(False)
df_1 = gen.sample(True)
for early_stop in [50, 40, 30, 20, 10]:
print("Running experiment A-bis with early stopping at an MSE of {:.1f}".format(early_stop))
hyperparams = MLP_HYPERPARAMS.copy()
hyperparams['early_stop'] = early_stop
# Experiment A-bis - Supports classification only
exp_class = pia.Experiment(gen, 'label', MLPEarlyStopping, N_TARGETS, N_SHADOWS, hyperparams, n_classes=2)
exp_class.run_targets()
exp_class.run_shadows()
res.loc[len(res)] = (early_stop,
exp_class.run_whitebox_deepsets(DEEPSETS_HYPERPARAMS, N_RUNS),
'Classification',
'WhiteBox')
res.loc[len(res)] = (early_stop,
exp_class.run_blackbox(N_RUNS),
'Classification',
'BlackBox')
# Since E(Y|X) changes depending on the classification label, we only test the models against their own class test set
for t, target in enumerate(exp_class.targets):
if exp_class.labels[t]:
perf.loc[len(perf)] = (early_stop, mean_squared_error(df_1.label, target.predict(df_1)), 'Classification')
else:
perf.loc[len(perf)] = (early_stop, mean_squared_error(df_0.label, target.predict(df_0)), 'Classification')
res.explode('MAE/Acc').reset_index(drop=True).to_pickle(join(results_folder, 'exp_abis_res.pkl'))
perf.to_pickle(join(results_folder, 'exp_abis_perf.pkl'))
def experiment_b(results_folder):
res = pd.DataFrame(columns=['early_stop', 'MAE/Acc', 'Experiment', 'Attack'])
perf = pd.DataFrame(columns=['early_stop', 'MSE', 'Experiment'])
gen = NN_Generator(2**11)
# Sampling target models' test sets
df_0 = gen.sample(False)
df_1 = gen.sample(True)
for early_stop in [50, 40, 30, 20, 10]:
print("Running experiment B with early stopping at an MSE of {:.1f}".format(early_stop))
hyperparams = MLP_HYPERPARAMS.copy()
hyperparams['early_stop'] = early_stop
# Experiment B - Regression
exp_reg = pia.Experiment(gen, 'label', MLPEarlyStopping, N_TARGETS, N_SHADOWS, hyperparams, n_classes=1, range=(0., 1.))
exp_reg.run_targets()
exp_reg.run_shadows()
res.loc[len(res)] = (early_stop,
exp_reg.run_whitebox_deepsets(DEEPSETS_HYPERPARAMS, N_RUNS),
'Regression',
'WhiteBox')
res.loc[len(res)] = (early_stop,
exp_reg.run_blackbox(N_RUNS),
'Regression',
'BlackBox')
# Experiment B - Classification
exp_class = pia.Experiment(gen, 'label', MLPEarlyStopping, N_TARGETS, N_SHADOWS, hyperparams, n_classes=2)
exp_class.run_targets()
exp_class.run_shadows()
res.loc[len(res)] = (early_stop,
exp_class.run_whitebox_deepsets(DEEPSETS_HYPERPARAMS, N_RUNS),
'Classification',
'WhiteBox')
res.loc[len(res)] = (early_stop,
exp_class.run_blackbox(N_RUNS),
'Classification',
'BlackBox')
# Target models performance
for t in exp_reg.targets:
perf.loc[len(perf)] = (early_stop, mean_squared_error(df_0.label, t.predict(df_0)), 'Regression')
perf.loc[len(perf)] = (early_stop, mean_squared_error(df_1.label, t.predict(df_1)), 'Regression')
for t in exp_class.targets:
perf.loc[len(perf)] = (early_stop, mean_squared_error(df_0.label, t.predict(df_0)), 'Classification')
perf.loc[len(perf)] = (early_stop, mean_squared_error(df_1.label, t.predict(df_1)), 'Classification')
res.explode('MAE/Acc').reset_index(drop=True).to_pickle(join(results_folder, 'exp_b_res.pkl'))
perf.to_pickle(join(results_folder, 'exp_b_perf.pkl'))
def experiment_c(results_folder):
res = pd.DataFrame(columns=['n_samples', 'MAE/Acc', 'Experiment', 'Attack'])
perf = pd.DataFrame(columns=['n_samples', 'MSE', 'Experiment'])
gen = NN_Generator(2**10)
# Sampling target models' test sets
df_0 = gen.sample(False)
df_1 = gen.sample(True)
for n_samples in range(512, 2049, 512):
print("Running experiment C with {} samples".format(n_samples))
gen.n_samples = n_samples
# Experiment C - Regression
exp_reg = pia.Experiment(gen, 'label', pia.MLP, N_TARGETS, N_SHADOWS, MLP_HYPERPARAMS, n_classes=1, range=(0., 1.), n_queries=512)
exp_reg.run_targets()
exp_reg.run_shadows()
res.loc[len(res)] = (n_samples,
exp_reg.run_whitebox_deepsets(DEEPSETS_HYPERPARAMS, N_RUNS),
'Regression',
'WhiteBox')
res.loc[len(res)] = (n_samples,
exp_reg.run_blackbox(N_RUNS),
'Regression',
'BlackBox')
# Experiment C - Classification
exp_class = pia.Experiment(gen, 'label', pia.MLP, N_TARGETS, N_SHADOWS, MLP_HYPERPARAMS, n_classes=2, n_queries=512)
exp_class.run_targets()
exp_class.run_shadows()
res.loc[len(res)] = (n_samples,
exp_class.run_whitebox_deepsets(DEEPSETS_HYPERPARAMS, N_RUNS),
'Classification',
'WhiteBox')
res.loc[len(res)] = (n_samples,
exp_class.run_blackbox(N_RUNS),
'Classification',
'BlackBox')
# Target models performance
for t in exp_reg.targets:
perf.loc[len(perf)] = (n_samples, mean_squared_error(df_0.label, t.predict(df_0)), 'Regression')
perf.loc[len(perf)] = (n_samples, mean_squared_error(df_1.label, t.predict(df_1)), 'Regression')
for t in exp_class.targets:
perf.loc[len(perf)] = (n_samples, mean_squared_error(df_0.label, t.predict(df_0)), 'Classification')
perf.loc[len(perf)] = (n_samples, mean_squared_error(df_1.label, t.predict(df_1)), 'Classification')
res.explode('MAE/Acc').reset_index(drop=True).to_pickle(join(results_folder, 'exp_c_res.pkl'))
perf.to_pickle(join(results_folder, 'exp_c_perf.pkl'))
def experiment_causal(results_folder):
res = pd.DataFrame(columns=['Model', 'MAE/Acc', 'Experiment', 'Attack'])
perf = pd.DataFrame(columns=['Model', 'Set', 'MSE', 'Experiment'])
# Test sets
gen = Membership_Generator()
df_0_valid = gen.sample(False)
df_1_valid = gen.sample(True)
df_0_test = deepcopy(df_0_valid)
df_0_test['x2'] = -df_0_test['x2']
df_1_test = deepcopy(df_1_valid)
df_1_test['x2'] = -df_1_test['x2']
for model_str in ['ERM', 'Causal ERM', 'IRM']:
# Experiment setup
print('Running causal experiment with model: ' + model_str)
gen = Membership_Generator(2**11, causal=(model_str == 'Causal ERM'), output_env=(model_str == 'IRM'))
hyperparams = {
'input_size': 2,
'n_classes': 1,
'epochs': 10,
'learning_rate': 1e-2,
'weight_decay': 1e-3,
'normalise': False,
'layers': (2,),
'bs': 256
}
if model_str == 'Causal ERM':
hyperparams['input_size'] = 1
elif model_str == 'IRM':
hyperparams['epochs'] = 2**11
hyperparams['reg'] = 0.1
hyperparams['env_label'] = 'env'
model = pia.MLP if not model_str == 'IRM' else IRM
# Causal - Classification
exp_class = pia.Experiment(gen, 'label', model, N_TARGETS, N_SHADOWS, hyperparams, n_classes=2)
exp_class.run_targets()
exp_class.run_shadows()
res.loc[len(res)] = (model_str,
exp_class.run_whitebox_deepsets(DEEPSETS_HYPERPARAMS, N_RUNS),
'Classification',
'WhiteBox')
res.loc[len(res)] = (model_str,
exp_class.run_blackbox(N_RUNS),
'Classification',
'BlackBox')
if model_str != 'Causal ERM':
for t in exp_class.targets:
perf.loc[len(perf)] = (model_str, 'Validation',
mean_squared_error(df_0_valid.label, t.predict(df_0_valid)), 'Classification')
perf.loc[len(perf)] = (model_str, 'Validation',
mean_squared_error(df_1_valid.label, t.predict(df_1_valid)), 'Classification')
perf.loc[len(perf)] = (model_str, 'Test',
mean_squared_error(df_0_test.label, t.predict(df_0_test)), 'Classification')
perf.loc[len(perf)] = (model_str, 'Test',
mean_squared_error(df_1_test.label, t.predict(df_1_test)), 'Classification')
else:
for t in exp_class.targets:
perf.loc[len(perf)] = (model_str, 'Validation',
mean_squared_error(df_0_valid.label, t.predict(df_0_valid.drop('x2', axis=1))), 'Classification')
perf.loc[len(perf)] = (model_str, 'Validation',
mean_squared_error(df_1_valid.label, t.predict(df_1_valid.drop('x2', axis=1))), 'Classification')
perf.loc[len(perf)] = (model_str, 'Test',
mean_squared_error(df_0_test.label, t.predict(df_0_test.drop('x2', axis=1))), 'Classification')
perf.loc[len(perf)] = (model_str, 'Test',
mean_squared_error(df_1_test.label, t.predict(df_1_test.drop('x2', axis=1))), 'Classification')
res.explode('MAE/Acc').reset_index(drop=True).to_pickle(join(results_folder, 'exp_causal_res.pkl'))
perf.to_pickle(join(results_folder, 'exp_causal_perf.pkl'))
if __name__ == '__main__':
if len(sys.argv) > 1:
exp = sys.argv[1]
res = 'results/'
if exp == 'a' or exp == 'A':
experiment_a(res)
elif exp == 'abis' or exp == 'A-bis':
experiment_abis(res)
elif exp == 'b' or exp == 'B':
experiment_b(res)
elif exp == 'c' or exp =='C':
experiment_c(res)
elif exp =='causal' or exp == 'd' or exp == 'D':
experiment_causal(res)