-
Notifications
You must be signed in to change notification settings - Fork 1
/
ad.py
243 lines (211 loc) · 10.1 KB
/
ad.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
from os import path, makedirs, getcwd, unlink, scandir
from joblib import *
from numpy import where, ravel, array_equal, flatnonzero, savetxt, unique
from sklearn.ensemble import IsolationForest
from sklearn.neighbors import LocalOutlierFactor
from sklearn.metrics import confusion_matrix, precision_recall_fscore_support
from sklearn.model_selection import GridSearchCV, RandomizedSearchCV
from yaml import safe_load, dump as ydump
from ad.utils import *
from sys import path as spath
spath.append(f"{spath[0]}/ad/thundersvm/python")
try:
from thundersvm import *
except ModuleNotFoundError:
svm_from_sk = False
from sklearn.svm import OneClassSVM
svm_from_sk = True
class AD(object):
"""docstring for AD"""
def __init__(self, blueprint, EXP_NAME='debug', n_jobs=1, verbose=True):
super(AD, self).__init__()
self.n_jobs = n_jobs
self.EXP_NAME = EXP_NAME
self.verbose = verbose
self.loaded = False
self.em, self.mv = -1, -1
if blueprint is None:
self.pseudo = True
self.model_type = f"pseudo"
else:
self.pseudo = False
self.from_save_model = True if '.jlib' in blueprint else False
if self.from_save_model:
print(f"{ACTION} loading trained model....")
if 'OneClassSVM' in blueprint and not svm_from_sk:
self.classifier = OneClassSVM()
self.classifier.load_from_file(blueprint)
else:
self.classifier = load(blueprint)
self.loaded = True
self.model_path = blueprint
else:
self.classifier = self._get_classifier(blueprint)
self.model_path = None
print(f"{'loaded' if self.loaded else 'build'} {self.__str__()}")
self.model_type = f"{self.classifier.__str__().split('(')[0]}"
salt = hash(self.classifier.get_params())[:8]
print(f"[*] model name: {BOLD}{salt}_{self.model_type}{CLR}")
def __str__(self):
return f"{self.classifier}@{self.classifier.get_params()}"
def fit_data(self, data, LOG_DIR=''):
start_time = time.now()
with console.status(f"fitting data..", spinner='dots') as status:
self.classifier.fit(data)
print(f"{ACTION} fitting took: {time.now()-start_time}")
if LOG_DIR != '':
AD_DIR = f"{LOG_DIR}/save_model/{self.EXP_NAME}/AD"
if not path.isdir(AD_DIR):
makedirs(AD_DIR)
params = self.classifier.get_params()
salt = hash(params)[:8]
print(f"{ACTION} saving {BOLD}{self.classifier}{CLR} classifier to -> {BOLD}file://{AD_DIR}{CLR}")
if 'OneClassSVM' in self.model_type and not svm_from_sk:
self.classifier.save_to_file(f"{AD_DIR}/{salt}_{self.model_type}_model.jlib")
else:
dump(self.classifier, f"{AD_DIR}/{salt}_{self.model_type}_model.jlib")
self.model_path = f"{AD_DIR}/{salt}_{self.model_type}_model.jlib"
with open(f"{AD_DIR}/{salt}_{self.model_type}_model.yaml", 'w') as outfile:
ydump({self.model_type: params}, outfile, default_flow_style=False)
def predict_data(self, data, em_mv=False):
start_time = time.now()
with console.status(f"predicting data..", spinner='dots') as status:
preds = self.classifier.predict(data)
print(f"{ACTION} predicting took: {time.now()-start_time}")
return preds
def grid_search(self, data, truth):
if self.model_type == 'OneClassSVM':
prefix = 'svm'
param_dist = {
"kernel": ['polynomial'],
"degree": [3, 5, 8, 11, 13, 15, 18],
"gamma": ['auto'],
"coef0": [0, 0.1, 0.5],
"tol": [0.001, 0.00001],
"nu": [0.5, 0.2, 0.001]
}
elif self.model_type == 'IsolationForest':
prefix = 'if'
param_dist = {
"n_estimators": [150, 100, 50],
"max_samples": [1, .5],
"max_features": [1, 1.5, 2],
"bootstrap": [True, False],
"contamination": [0, 0.25 ,0.5]
}
elif self.model_type == 'LocalOutlierFactor':
prefix = 'lof'
param_dist = {
"n_neighbors": [40 ,20, 10],
"algorithm": ['auto', 'ball_tree', 'kd_tree', 'brute'],
"leaf_size": [10, 20, 40],
"p": [1, 1.5, 2],
}
WRITER = spawn_tensorboard(prefix)
n_iter_search = 20
random_search = GridSearchCV(self.classifier,
param_dist,
scoring=f1_scorer,)
#n_iter=n_iter_search)
start_time = time.now()
random_search.fit(data, get_gt(truth))
print(f"[*] grid search took: {time.now()-start_time} for {n_iter_search} candidates parameter settings.")
results = random_search.cv_results_
n_top = 5
for i in range(1, n_top + 1):
candidates = flatnonzero(results['rank_test_score'] == i)
for candidate in candidates:
print(f"Model with rank: {i}")
print(f"Mean validation score: {results['mean_test_score'][candidate]:.3f} (std: {results['std_test_score'][candidate]:.3f})")
print(f"Parameters: {results['params'][candidate]}\n")
WRITER.add_hparams(hparam_dict=results['params'][candidate],
metric_dict={f'hparam/f1':results['mean_test_score'][candidate]},
run_name=f"{self.EXP_NAME}_{i}")
def clean(self, log_dir):
buffer = [fd.path for fd in scandir(f"{log_dir}") if 'tmp' in fd.path]
for file in buffer:
unlink(file)
print(f"[*] cleand {len(buffer)} files")
def calc_metrics(self, predicted, truth, show=True, save='', WRITER=None, **kwargs):
log_dir = kwargs['log_dir']
truth_path = None
if isinstance(truth, str):
truth_path = truth
truth = get_gt(truth)
num_of_preds = len(predicted)
assert num_of_preds == len(truth), exit(f"{ACTION} number of predicts {num_of_preds} does not match number of labels {len(truth)}")
threshold = kwargs['threshold'] if 'threshold' in kwargs.keys() else None
time = f"{kwargs['time']}" if 'time' in kwargs.keys() else "N/A"
# calculate metrics
metric_dict = get_metrics(truth, predicted, show=False, NORMAL_LABEL=NORMAL, ANOMALY_LABEL=ANOMALY)
if truth_path is not None:
del truth
TN = metric_dict['TN']
FP = metric_dict['FP']
FN = metric_dict['FN']
TP = metric_dict['TP']
pr = metric_dict['pr']
re = metric_dict['re']
f1 = metric_dict['f1']
fpr = metric_dict['fpr']
if show:
plot_cm_to_console(TN=TN, FP=FP, FN=FN, TP=TP, EXP_NAME=f"{self.EXP_NAME}{save}")
plot_metrics_to_console(pr=pr, re=re, f1=f1, fpr=fpr, em=self.em, mv=self.mv, EXP_NAME=self.EXP_NAME)
if save != '':
hash_ = hash(self.classifier.get_params()) if not self.pseudo else hash("dummy")
blob = {'hash': hash_[:8],
'type': self.model_type,
'name': self.EXP_NAME,
'data path': save,
'threshold': threshold,
'TN': TN,
'FP': FP,
'FN': FN,
'TP': TP,
'FPR': fpr,
'PR': pr,
'RE': re,
'f1': f1,
'em': self.em,
'mv': self.mv,
'time':time,
'model path': self.model_path if not self.pseudo else "pcapAE"}
save_results(save_path=f"{log_dir}/logs/results.csv", **blob)
print(f"{ACTION} adding row to {BOLD}{log_dir}/logs/results.csv{CLR}")
if 'save_ano_pids' in kwargs.keys():
tmp_path = f"ano_pids_{kwargs['in_file'].replace('/','_')}_{self.EXP_NAME.split('*')[0].replace('/','_')}_{self.model_type.replace('/','_')}"
if not path.isdir(f'{log_dir}/logs/{tmp_path}'):
makedirs(f'{log_dir}/logs/{tmp_path}')
print(f"{ACTION} saving confusion matrix PIDs to {BOLD}runs/logs/{tmp_path}{CLR}")
if truth_path is not None:
ano_pids = get_ano_pids(get_gt(truth_path, skip=True), predicted)
else:
ano_pids = get_ano_pids(truth, predicted)
del truth
for case, data in zip(['TP','FP','FN'], ano_pids):
save_file = f"{log_dir}/logs/{tmp_path}/{case}.txt"
savetxt(save_file,
data,
fmt='%s')
if WRITER is not None:
WRITER.add_text(f"AD_{self.model_type}_{self.EXP_NAME}", str(blob), global_step=None, walltime=None)
def _get_classifier(self, blueprint):
with open(blueprint, 'r') as stream:
blueprint = safe_load(stream)
model = list(blueprint.keys())[0]
params = blueprint[model]
if model == 'OneClassSVM':
estimator = OneClassSVM()
params['verbose'] = self.verbose
elif model == 'IsolationForest':
estimator = IsolationForest()
params['verbose'] = self.verbose
params['n_jobs'] = self.n_jobs
params['random_state'] = 42
elif model == 'LocalOutlierFactor':
estimator = LocalOutlierFactor()
params['n_jobs'] = self.n_jobs
else:
raise(f"NYI\n{blueprint}")
estimator.set_params(**params)
return estimator