-
Notifications
You must be signed in to change notification settings - Fork 0
/
biasattack.py
179 lines (154 loc) · 5.97 KB
/
biasattack.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
import numpy as np
import random
import torch
import textattack
from textattack.shared.attack import Attack
from textattack.transformations import CompositeTransformation, WordDeletion
import utils
from constraints import WordBlacklistConstraint, ActiveAnchorWordsModification, LanguageModelScoreConstraint
from goal_function import BiasGoalFunctionResult, BiasGoalFunction
from search_method import BeamSearchPlus
from transformations import Identity, InitialBiasWord, WordSwapMaskedLMPlus
def BiasAttack(model,
score_mode,
biaswords_list=None,
biaswords_type='counterfitted',
biasthreshold=0.5,
diffthreshold=0.03,
max_masks=1,
max_trials=-1,
beam_width=5,
max_candidates=20,
logit_threshold=3,
beam_sampling_method='max',
transformations_sampling_ratio=1,
search_depth=100,
query_budget=5000,
maximizable=False):
"""
BiasAttack main entrance.
"""
np.set_printoptions(precision=6)
if biaswords_list is None:
# biaswords_list, bias_urls = utils.get_gender_swap_words()
biaswords_list = utils.get_anchor_words(model, gen_type=biaswords_type)
print(
f"Loaded {len(biaswords_list)} biaswords from utils.\n Samples: {biaswords_list[:10]}...{biaswords_list[-10:]}"
)
else:
print(
f"Loaded {len(biaswords_list)} biaswords from command line args.\n Samples: {biaswords_list[:10]}"
)
biaswords_flatten = utils.flatten_nested_list(biaswords_list)
transformation = CompositeTransformation([
InitialBiasWord(biaswords_flatten=biaswords_flatten),
WordSwapMaskedLMPlus(max_masks=max_masks,
max_trials=max_trials,
max_candidates=max_candidates,
logit_threshold=logit_threshold,
max_length=128,
masked_language_model="distilbert-base-uncased"),
])
blacklist_words = [*biaswords_flatten]
gender_define_words, gender_define_urls = utils.get_gender_define_words()
blacklist_words.extend(gender_define_words)
blacklist_words = [w.strip() for w in blacklist_words]
print(
f"Loaded {len(blacklist_words)} blacklist_words from {gender_define_urls}.\n Samples: {blacklist_words[:10]}...{blacklist_words[-10:]}"
)
constraints = [
ActiveAnchorWordsModification(),
]
# Note: Each |query_budget| corresponds to |len(biaswords)| actual model queries.
gpu_memory_MB = torch.cuda.get_device_properties(
textattack.shared.utils.device).total_memory // (2**20)
print(f'gpu_memory_MB = {gpu_memory_MB}')
if gpu_memory_MB >= 8000:
model_batch_size = 512
else:
model_batch_size = 128
goal_function = BiasGoalFunction(model,
maximizable=maximizable,
biaswords_list=biaswords_list,
active_biaswords_logit_threshold=1.5,
biasthreshold=biasthreshold,
diffthreshold=diffthreshold,
stepweight=0.1,
score_mode=score_mode,
lm_scorer="distilbert-base-uncased",
query_budget=query_budget,
model_batch_size=model_batch_size)
search_method = BeamSearchPlus(
beam_width=beam_width,
beam_sampling_method=beam_sampling_method,
transformations_sampling_ratio=transformations_sampling_ratio,
search_depth=search_depth)
return Attack(goal_function, constraints, transformation, search_method)
def SOEnumAttack(model, biaswords_str):
return BiasAttack(
model,
score_mode="max",
biaswords_type='counterfitted',
biaswords_list=utils.biaswords_list_from_str(biaswords_str),
beam_width=2000,
search_depth=2,
query_budget=100000)
def SOBeamAttack(model, biaswords_str):
return BiasAttack(
model,
score_mode="max",
biaswords_type='counterfitted',
biaswords_list=utils.biaswords_list_from_str(biaswords_str),
max_masks=1,
max_trials=-1,
beam_width=20,
search_depth=6,
query_budget=50000)
def RandomBaselineAttack(model, biaswords_str):
return BiasAttack(
model,
score_mode="max",
beam_sampling_method="random",
biaswords_type='counterfitted',
biaswords_list=utils.biaswords_list_from_str(biaswords_str),
max_masks=1,
max_trials=1,
beam_width=1,
search_depth=6,
query_budget=100)
def RandomNoLMBaselineAttack(model, biaswords_str):
return BiasAttack(
model,
score_mode="max",
beam_sampling_method="random",
biaswords_type='counterfitted',
biaswords_list=utils.biaswords_list_from_str(biaswords_str),
max_candidates=1000,
logit_threshold=10,
max_masks=1,
max_trials=1,
beam_width=1,
search_depth=6,
query_budget=100)
def BiasAnalysisChecklist(model, biaswords_str):
return BiasAttack(
model,
score_mode="max",
biaswords_type='checklist',
biaswords_list=utils.biaswords_list_from_str(biaswords_str),
beam_sampling_method='random',
beam_width=800,
search_depth=3,
query_budget=100000,
maximizable=True)
def BiasAnalysisGender(model, biaswords_str):
return BiasAttack(
model,
score_mode="max",
biaswords_type='gender',
biaswords_list=utils.biaswords_list_from_str(biaswords_str),
beam_sampling_method='random',
beam_width=800,
search_depth=3,
query_budget=100000,
maximizable=True)