-
Notifications
You must be signed in to change notification settings - Fork 1
/
majority_smooth.py
252 lines (219 loc) · 9.55 KB
/
majority_smooth.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
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.cluster import KMeans
from utils import *
import re
from collections import Counter
import argparse
from openai_api import keyword_extract
def read_file(file_path):
with open(file_path, 'r') as file:
lines = file.readlines()
return [line.strip() for line in lines]
def cluster_kmeans(sentences, num_clusters=2):
# Vectorize the sentences using TF-IDF
vectorizer = TfidfVectorizer(stop_words='english')
X = vectorizer.fit_transform(sentences)
# Apply KMeans clustering
kmeans = KMeans(n_clusters=num_clusters, random_state=0).fit(X)
return kmeans.cluster_centers_, kmeans.labels_
def anomaly_keywords(rule_path = 'rule/rule_SHTech.txt', regenerate_keyword = False):
'''
The below anomaly keywords are extracted once and used for the experiment in the paper,
you can also extract from your rules with the below function.
'''
if regenerate_keyword == False:
anomaly_from_rule = [
"trolley",
"cart",
"luggage",
"bicycle",
"skateboard",
"scooter",
"vehicles",
"vans",
"accident",
"running",
"jumping",
"riding",
"skateboarding",
"scooting",
"lying",
"falling",
"bending",
"fighting",
"pushing",
"loitering",
"climbing",
"tampering",
"lingering"]
else:
anomaly_from_rule = keyword_extract(rule_path)
return anomaly_from_rule
def cluster_keyword(text_lines, anomaly_from_rule):
preds = []
anomaly_word = []
for line in text_lines:
found_anomaly = False
for anomaly in anomaly_from_rule:
if anomaly in line:
found_anomaly = True
anomaly_word.append(anomaly)
preds.append(1 if found_anomaly else 0)
return preds, anomaly_from_rule, anomaly_word
def majority_smooth(data, window_size=20, edge_region_size=None):
# Adjust window size to be odd
if window_size % 2 == 0:
window_size += 1
pad_size = window_size // 2
# Determine edge region size if not specified
if edge_region_size is None:
edge_region_size = pad_size
padded_data = np.pad(data, pad_size, mode='edge')
smoothed_data = np.copy(data)
for i in range(len(data)):
# Apply different rule for edge regions
if i < edge_region_size or i >= len(data) - edge_region_size:
# For edge data, consider only the previous pad_size values
start = max(0, i - pad_size)
end = i + 1 # Include the current point
window = padded_data[start:end]
else:
# Regular processing for central part
start = i
end = i + window_sizez
window = padded_data[start:end]
# Apply majority rule
ones_count = np.sum(window)
zeros_count = len(window) - ones_count
smoothed_data[i] = 1 if ones_count > zeros_count else 0
return smoothed_data
def ema_majority_smooth(ema_data, threshold, window_size=20, edge_region_size=None):
# Adjust window size to be odd
if window_size % 2 == 0:
window_size += 1
pad_size = window_size // 2
# Determine edge region size if not specified
if edge_region_size is None:
edge_region_size = pad_size
padded_data = np.pad(ema_data, pad_size, mode='edge')
smoothed_data = np.zeros(len(ema_data), dtype=int)
for i in range(len(ema_data)):
# Apply different rule for edge regions
if i < edge_region_size or i >= len(ema_data) - edge_region_size:
# For edge data, consider only the previous pad_size values
start = max(0, i - pad_size)
end = i + 1 # Include the current point
window = padded_data[start:end]
else:
# Regular processing for central part
start = i
end = i + window_size
window = padded_data[start:end]
# Apply majority rule based on threshold
above_threshold_count = np.sum(window > threshold)
below_threshold_count = len(window) - above_threshold_count
smoothed_data[i] = 1 if above_threshold_count > below_threshold_count else 0
return smoothed_data
def find_most_frequent_keyword(text, keyword_list):
words = re.findall(r'\b\w+\b', text)
keyword_freq = Counter(word for word in words if word in keyword_list)
if keyword_freq:
return keyword_freq.most_common(1)[0][0]
return None
def remove_sentences_with_keywords(text, keyword_list):
# Splitting the text into partial sentences
partial_sentences = re.split(r',|\.', text)
# Keeping sentences that do not contain any of the keywords
return '. '.join(sentence for sentence in partial_sentences if not any(keyword in sentence for keyword in keyword_list))
def modify_text(preds, s_preds, keyword_list, text_list, window_size):
if window_size % 2 != 0:
window_size += 1
window_size = int(window_size/2)
modified_text_list = []
for i, (pred, s_pred, text) in enumerate(zip(preds, s_preds, text_list)):
# Condition when original label is 1 and new label is 0
if s_pred == 0:
if pred == 1:
text = remove_sentences_with_keywords(text, keyword_list)
# Condition when new label is 1
elif s_pred == 1:
# Extracting the window of text
start_index = max(0, i - window_size)
end_index = min(len(text_list), i + window_size + 1)
window_text = ' '.join(text_list[start_index:end_index])
most_freq_keyword = find_most_frequent_keyword(window_text, keyword_list)
if most_freq_keyword:
if most_freq_keyword.endswith('ing'):
addition = f'{most_freq_keyword}'
else:
addition = f'riding a {most_freq_keyword}'
# text = re.sub(r'\.', f'. {addition}', text, 1)
pattern = r"(the first person is)[^,]*"
text = re.sub(pattern, r"\1 " + addition, text)
modified_text_list.append(text)
return modified_text_list
def evaluate(file_path, labels, output_file_path, save_modified,anomaly_from_rule):
# Initial labels using keyword in rules
text_lines = read_file(file_path)
preds, keyword_list, _ = cluster_keyword(text_lines, anomaly_from_rule=anomaly_from_rule)
# First-time EMA to smooth the preds with a more sensitive way
ema_smoothed_data = pd.Series(preds).ewm(alpha = 0.33, adjust=True).mean()
threshold = ema_smoothed_data.mean()
s_preds = ema_majority_smooth(ema_smoothed_data, threshold, window_size=1)
if threshold ==0:
threshold += 0.0000001
# Second-time EMA to get the auc score
scores = pd.Series(s_preds).ewm(alpha = threshold, adjust=True).mean()
if save_modified == True:
modified_texts = modify_text(preds, s_preds, keyword_list, text_lines, window_size=1)
with open(output_file_path, 'w') as file:
for inner_list in modified_texts:
file.write(inner_list + '\n')
print(f"======================{file_path.split('/')[-1].split('.')[0]}========================> ")
print(f'Ori ACC: {accuracy_score(labels, preds)}')
print(f'Ori Precision: {precision_score(labels, preds)}')
print(f'Ori Recall: {recall_score(labels, preds)}')
print(f'Soomth ACC: {accuracy_score(labels, s_preds)}')
print(f'Soomth Precision: {precision_score(labels, s_preds)}')
print(f'Soomth Recall: {recall_score(labels, s_preds)}')
return preds, list(s_preds), list(scores), list(ema_smoothed_data)
def parse_arguments():
parser = argparse.ArgumentParser()
parser.add_argument('--data', type=str, default='SHTech',
choices=['SHTech', 'avenue', 'ped2', 'UBNormal'])
args = parser.parse_args()
return args
def main():
args = parse_arguments()
data_name = args.data
entries = os.listdir(f'{data_name}/test_frame_description')
all_preds = []
all_labels = []
all_spreds = []
all_scores = []
all_ori_scores = []
anomaly_from_rule = anomaly_keywords(rule_path='rule/rule_SHTech.txt')
for item in entries:
name = item.split('.')[0]
input_file_path = f'{data_name}/test_frame_description/{name}.txt' # Path to your input text file
output_file_path = f'{data_name}/modified_test_frame_description/{name}.txt' # Path for the new output text file
if not os.path.exists(os.path.dirname(output_file_path)):
os.makedirs(os.path.dirname(output_file_path))
labels = pd.read_csv(f'{data_name}/test_frame/{name}.csv').iloc[:, 1].tolist()
preds, s_preds, scores, ori_scores = evaluate(input_file_path, labels, output_file_path, save_modified=False, anomaly_from_rule=anomaly_from_rule)
all_labels += labels
all_preds += preds
all_spreds += s_preds
all_scores += scores
all_ori_scores += ori_scores
print(f"======================ALL DATA========================> ")
print(f'Ori ACC: {accuracy_score(all_labels, all_preds)}')
print(f'Ori Precision: {precision_score(all_labels, all_preds)}')
print(f'Ori Recall: {recall_score(all_labels, all_preds)}')
print(f'Ori AUC: {roc_auc_score(all_labels, all_ori_scores)}')
print(f'Smooth ACC: {accuracy_score(all_labels, all_spreds)}')
print(f'Smooth Precision: {precision_score(all_labels, all_spreds)}')
print(f'Smooth Recall: {recall_score(all_labels, all_spreds)}')
print(f'AUC: {roc_auc_score(all_labels, all_scores)}')
if __name__ == "__main__":
main()