def load_sequence_data_by_tshark(first_n_pkts_input_file, separator=','):
"""
:param first_n_pkts_input_file: E.g. input_file = '../results/AUDIO_first_n_pkts_10_all_in_one_file.txt'
:param separator:
:return: X=Features, Y=Label
"""
data = []
label = []
with open(first_n_pkts_input_file, 'r') as fid_in:
line = fid_in.readline()
while line:
# No, time, srcIP dstIP, protocol, pkts_size, srcPort, dstPort
line_arr = line.split()
if len(line_arr) < 9:
print('skip: ', line[:-2]) # reomve '\n'
line = fid_in.readline()
continue
data.append([line_arr[-3], line_arr[-2]])
# print([line_arr[-3], line_arr[-3]])
label.append(line_arr[-1].split('\n')[0])
line = fid_in.readline()
Y = change_label(label)
new_data = normalize_data(np.asarray(data, dtype=float),
range_value=[0, 1],
eps=1e-5)
X = []
for idx in range(len(new_data)):
line_arr = new_data[idx]
# len_tmp = int(line_arr[4]) # length of pkts_list
line_tmp = []
# for i in range(1, len_tmp + 1): # len(pkts_list), [1, len_tmp+1)
# if i == 1:
# line_tmp.append([line_arr[0],line_arr[1],line_arr[2], line_arr[3], line_arr[4 + i], line_arr[4 + len_tmp + i]]) # srcport, dstport, [pkts_lst[0], flow_duration]
# else:
# line_tmp.append([line_arr[0],line_arr[1],line_arr[2], line_arr[3],line_arr[4 + i], line_arr[
# 4 + len_tmp + (i + 1)]]) # [pkts_lst[0], intr_tm_lst[1]], intr_tm_lst from 1, 2, ...
# line_tmp=[line_arr[-3], line_arr[-3]] # pkts_len
line_tmp = [line_arr]
X.append(line_tmp)
return X, Y
def __init__(self, input_file, transform=None, normalization_flg=False):
self.X = []
self.y = []
with open(input_file, 'r') as fid_in:
line = fid_in.readline()
while line:
line_arr = line.split(',')
value = list(map(lambda x: float(x), line_arr[:-1]))
self.X.append(value)
self.y.append(float(line_arr[-1].strip()))
line = fid_in.readline()
if normalization_flg:
self.X = normalize_data(np.asarray(self.X, dtype=float),
range_value=[-1, 1],
eps=1e-5)
self.transform = transform
def load_sequence_data_backup(first_n_pkts_input_file, separator=','):
# input_file = '../results/AUDIO_first_n_pkts_10_all_in_one_file.txt'
data = []
label = []
with open(first_n_pkts_input_file, 'r') as fid_in:
line = fid_in.readline()
while line:
### srcIP, dstIP, srcport, dstport, len(pkts), pkts_lst, flow_duration, intr_time_lst, label
line_arr = line.split(separator)
len_tmp = int(line_arr[4]) # length of pkts_list
data.append(line_arr[:-1])
label.append(line_arr[-1].split('\n')[0])
line = fid_in.readline()
# X = normalize_data(np.asarray(X, dtype=float), range_value=[0, 1], eps=1e-5)
Y = change_label(label)
new_data = normalize_data(np.asarray(data, dtype=float),
range_value=[0, 1],
eps=1e-5)
X = []
for idx in range(len(new_data)):
line_arr = new_data[idx]
# len_tmp = int(line_arr[4]) # length of pkts_list
line_tmp = []
for i in range(1, len_tmp + 1): # len(pkts_list), [1, len_tmp+1)
if i == 1:
line_tmp.append([
line_arr[0], line_arr[1], line_arr[2], line_arr[3],
line_arr[4 + i], line_arr[4 + len_tmp + i]
]) # srcport, dstport, [pkts_lst[0], flow_duration]
else:
line_tmp.append([
line_arr[0], line_arr[1], line_arr[2], line_arr[3],
line_arr[4 + i], line_arr[4 + len_tmp + (i + 1)]
]) # [pkts_lst[0], intr_tm_lst[1]], intr_tm_lst from 1, 2, ...
X.append(line_tmp)
return X, Y
label_integer = np.asarray(label_integer, dtype=int)
onehot_encoder = OneHotEncoder(sparse=False)
integer_encoded = label_integer.reshape(len(label_integer), 1)
onehot_encoded = onehot_encoder.fit_transform(integer_encoded)
return np.array(onehot_encoded, dtype=int)
if __name__ == '__main__':
torch.manual_seed(1) # reproducible
input_file = '../results/AUDIO_first_n_pkts_10_all_in_one_file.txt'
# X, Y = load_data(input_file)
X, Y = load_data_compute_mean(input_file)
X = normalize_data(np.asarray(X, dtype=float),
range_value=[0, 1],
eps=1e-5)
Y = change_label(Y)
X_train, X_test, y_train, y_test = achieve_train_test_data(X,
Y,
train_size=0.9,
shuffle=True)
ann = MLP(BATCH_SIZE=20,
first_n_pkts=10,
epochs=10,
num_class=len(Counter(y_train)))
# training_set = Data.TensorDataset(torch.Tensor(X_train), torch.Tensor(y_train)) # X, Y
one_hot_y_train = one_hot_sklearn(y_train)
training_set = (X_train, y_train)
ann.train(training_set)