-
Notifications
You must be signed in to change notification settings - Fork 6
/
wd_kfold_unsw.py
304 lines (258 loc) · 11.5 KB
/
wd_kfold_unsw.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
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from sklearn.preprocessing import MinMaxScaler, QuantileTransformer
from sklearn.model_selection import StratifiedKFold
from preprocess.unsw import get_feature_names, discovery_feature_volcabulary
from preprocess.unsw import generate_header, discovery_discrete_range
from netlearner.utils import measure_prediction
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
import tensorflow as tf
import pickle
import os
def test_splitted_data_builder(train_path, test_path):
for i in range(fold):
train_fold = train_path + '.train_fold%d.csv' % i
valid_fold = train_path + '.valid_fold%d.csv' % i
train_ib, _ = input_builder(train_fold, columns)
valid_ib, _ = input_builder(valid_fold, columns)
test_ib, ohe = input_builder(test_path, columns)
train_ib, _ = input_builder(train_path, columns)
def plot_history(train_loss, valid_loss, test_loss, fig_dir):
fig, ax1 = plt.subplots()
ln1 = ax1.plot(train_loss, 'r--', label='Train')
ln2 = ax1.plot(valid_loss, 'b:', label='Valid')
ax1.set_ylabel('Train/Valid Loss', color='r')
ax2 = ax1.twinx()
ln3 = ax2.plot(test_loss, 'g-.', label='Test')
ax2.set_ylabel('Test Loss', color='g')
lns = ln1 + ln2 + ln3
labels = [l.get_label() for l in lns]
ax1.legend(lns, labels, loc='upper left')
ax1.grid(color='k', linestyle=':', linewidth=1)
ax2.grid(color='k', linestyle=':', linewidth=1)
fig.tight_layout()
plt.savefig(fig_dir + 'history.pdf', format='pdf')
plt.close()
def build_model(model_dir):
hidden_layers = [800, 480]
label_names = label_map.keys()
m = tf.estimator.DNNLinearCombinedClassifier(
model_dir=model_dir,
linear_feature_columns=wide_columns,
dnn_feature_columns=deep_columns,
dnn_hidden_units=hidden_layers,
dnn_dropout=dropout,
label_vocabulary=label_names,
n_classes=len(label_names))
print('Hidden units in each layer:%s' % hidden_layers)
return m
def process_dataset(filename, output_path, split):
global scaler_fitted, transformer_fitted
print('Process %s' % filename)
df = pd.read_csv(filename, names=CSV_COLUMNS, sep=',',
skipinitialspace=True, skiprows=1, engine='python')
labels = df['label'] if binary_class else df['attack_cat']
df.drop(['label', 'attack_cat'], axis=1)
numeric = df[continuous_names + discrete_names].as_matrix()
symbolic = df[symbolic_names].as_matrix()
if scaler_fitted is False:
scaler.fit(numeric)
scaler_fitted = True
normalized = scaler.transform(numeric)
full_columns = symbolic_names + continuous_names + discrete_names
combined = np.concatenate((symbolic, normalized), axis=1)
if len(quantile_names) > 0:
if transformer_fitted is False:
transformer.fit(numeric)
transformer_fitted = True
augment = transformer.transform(numeric)
combined = np.concatenate((combined, augment), axis=1)
full_columns += quantile_names
print('Raw dataset shape', combined.shape)
print('Raw label shape', labels.shape)
if split is True:
skf = StratifiedKFold(n_splits=fold)
X = combined.copy()
y = np.reshape(labels, (-1, 1))
i = 0
for train_index, valid_index in skf.split(X, labels):
train_dataset, valid_dataset = X[train_index], X[valid_index]
train_labels, valid_labels = y[train_index], y[valid_index]
train = np.concatenate((train_dataset, train_labels), axis=1)
train_df = pd.DataFrame(train, columns=full_columns + ['label'])
train_df.to_csv(output_path + '.train_fold%d.csv' % i, index=False)
valid = np.concatenate((valid_dataset, valid_labels), axis=1)
valid_df = pd.DataFrame(valid, columns=full_columns + ['label'])
valid_df.to_csv(output_path + '.valid_fold%d.csv' % i, index=False)
i += 1
combined_df = pd.DataFrame(combined, columns=full_columns,
index=labels.index.tolist())
save = pd.concat([combined_df, labels], axis=1)
save.to_csv(output_path, index=False)
return full_columns + ['label']
def input_builder(filename, full_columns):
print('Building input for %s' % filename)
dtypes = dict(zip(discrete_names, [np.int32] * len(discrete_names)))
df = pd.read_csv(filename, names=full_columns, sep=',',
skipinitialspace=True, skiprows=1,
engine='python', dtype=dtypes)
labels = df['label'].astype(str)
labels_ohe = np.zeros((labels.shape[0], len(label_map)))
for (i, x) in enumerate(labels):
labels_ohe[i][label_map[x]] = 1.0
dataset = df.drop('label', axis=1)
print('Dataset shape:', dataset.shape)
print('Label shape:', labels.shape)
ib = tf.estimator.inputs.pandas_input_fn(dataset, labels, batch_size,
shuffle=True, num_threads=1)
return ib, labels_ohe
def train_and_eval(model_dir, train_path, test_path, columns):
hist = {'train_loss': [], 'valid_loss': [],
'test_loss': [], 'test_acc': []}
fold_train_loss, fold_valid_loss = [], []
for i in range(fold):
train_fold = train_path + '.train_fold%d.csv' % i
valid_fold = train_path + '.valid_fold%d.csv' % i
m = build_model(model_dir=None)
train_loss, valid_loss = [], []
for _ in range(num_epochs):
train_ib, _ = input_builder(train_fold, columns)
valid_ib, _ = input_builder(valid_fold, columns)
m.train(train_ib)
scores = m.evaluate(input_fn=train_ib)
train_loss.append(scores['average_loss'])
scores = m.evaluate(input_fn=valid_ib)
valid_loss.append(scores['average_loss'])
fold_train_loss.append(train_loss)
fold_valid_loss.append(valid_loss)
print('Train fold %d finished' % i)
hist['train_loss'] = np.mean(fold_train_loss, axis=0)
hist['valid_loss'] = np.mean(fold_valid_loss, axis=0)
opt_epochs = np.argmin(hist['valid_loss'])
print('Optimal #Epochs:', opt_epochs + 1)
hist['opt_epochs'] = opt_epochs + 1
test_ib, ohe = input_builder(test_path, columns)
m = build_model(None)
for i in range(num_epochs):
train_ib, _ = input_builder(train_path, columns)
m.train(input_fn=train_ib)
scores = m.evaluate(test_ib)
hist['test_loss'].append(scores['average_loss'])
hist['test_acc'].append(scores['accuracy'])
hist['test_acc_report'] = hist['test_acc'][opt_epochs]
print('Test accuracy = %s' % hist['test_acc_report'])
predictions = np.zeros_like(ohe)
for (i, x) in enumerate(list(m.predict(test_ib))):
predictions[i][x['class_ids'][0]] = 1.0
conf_table = measure_prediction(np.array(predictions), ohe, model_dir)
hist['confusion_table'] = conf_table
print(conf_table)
return hist
os.environ['CUDA_VISIBLE_DEVICES'] = '3'
train_filename = 'UNSW/UNSW_NB15_training-set.csv'
test_filename = 'UNSW/UNSW_NB15_testing-set.csv'
feature_filename = 'UNSW/feature_names_train_test.csv'
CSV_COLUMNS, symbolic_names, continuous_names, discrete_names = \
get_feature_names(feature_filename)
upper, lower, small_ranges = discovery_discrete_range(
[train_filename, test_filename], discrete_names, CSV_COLUMNS)
quantile_names = []
for name in continuous_names + discrete_names:
quantile_names.append(name + '_quantile')
print(symbolic_names, len(symbolic_names))
print(continuous_names, len(continuous_names))
print(discrete_names, len(discrete_names))
print(quantile_names, len(quantile_names))
header = generate_header(CSV_COLUMNS)
print('Headers for dataset file:', header)
symbolic_features = discovery_feature_volcabulary([train_filename,
test_filename])
symbolic_columns = dict()
for (name, categorical_values) in symbolic_features.items():
column = tf.feature_column.categorical_column_with_vocabulary_list(
name, categorical_values)
symbolic_columns[name] = column
continuous_columns = dict()
for name in continuous_names + discrete_names + quantile_names:
column = tf.feature_column.numeric_column(name)
continuous_columns[name] = column
# convert discrete features into categorical columns
discrete_columns = dict()
for name in small_ranges:
column = tf.feature_column.categorical_column_with_identity(
name, num_buckets=upper[name] - lower[name] + 1)
discrete_columns[name] = column
# Build components for the wide model
base_columns = symbolic_columns.values() + discrete_columns.values()
cross_columns = [
tf.feature_column.crossed_column(
['proto', 'service'], hash_bucket_size=1000),
tf.feature_column.crossed_column(
['proto', 'state'], hash_bucket_size=1000),
tf.feature_column.crossed_column(
['service', 'state'], hash_bucket_size=200),
tf.feature_column.crossed_column(
['proto', 'service', 'state'], hash_bucket_size=1600)
]
wide_columns = base_columns + cross_columns
print('#wide components:', len(wide_columns))
# Build components for the deep model
indicator_columns = [] # low dimension categorical features
for name in ['state', 'service']:
column = symbolic_columns[name]
indicator_columns.append(tf.feature_column.indicator_column(column))
print('indicator columns', len(indicator_columns))
"""
low_discrete_names = ['trans_depth', 'ct_state_ttl',
'ct_flw_http_mthd', 'ct_ftp_cmd']
for name in low_discrete_names:
column = discrete_columns[name]
indicator_columns.append(tf.feature_column.indicator_column(column))
"""
# convert high dimension categorical features to embeddings
embedding_columns = []
for (name, column) in symbolic_columns.items():
volcabulary_size = len(symbolic_features[name])
# print(name, '|V| =', volcabulary_size)
dim = np.ceil(np.log2(volcabulary_size))
print('Embedding size of %s is %d' % (name, dim))
embedding = tf.feature_column.embedding_column(column, dim)
embedding_columns.append(embedding)
for (name, column) in discrete_columns.items():
volcabulary_size = upper[name] - lower[name] + 1
# print(name, '|V| =', volcabulary_size)
dim = np.ceil(np.log2(volcabulary_size))
print('Embedding size of %s is %d' % (name, dim))
embedding = tf.feature_column.embedding_column(column, int(dim))
embedding_columns.append(embedding)
print('embedding columns', len(embedding_columns))
deep_columns = indicator_columns + embedding_columns \
+ continuous_columns.values()
print('#deep components:', len(deep_columns))
model_dir = 'WideDeepModel/UNSW/'
train_path = model_dir + 'aug_train.csv'
test_path = model_dir + 'aug_test.csv'
num_epochs = 400
batch_size = 64
dropout = 0.2
fold = 5
transformer = QuantileTransformer()
transformer_fitted = False
scaler = MinMaxScaler()
scaler_fitted = False
label_map = {'0': 0, '1': 1}
# label_map = {"Normal": 0, "Backdoor": 1, "Analysis": 2, "Fuzzers": 3,
# "Reconnaissance": 4, "Exploits": 5, "DoS": 6,
# "Shellcode": 7, "Worms": 8, "Generic": 9}
binary_class = (len(label_map) == 2)
columns = process_dataset(train_filename, train_path, split=True)
process_dataset(test_filename, test_path, split=False)
hist = train_and_eval(model_dir, train_path, test_path, columns)
plot_history(hist['train_loss'], hist['valid_loss'],
hist['test_loss'], model_dir)
output = open(model_dir + 'Runs%d.pkl' % (num_epochs), 'wb')
pickle.dump(hist, output)
output.close()