def _get_data(data_pth=None, data_name=None, direction='src_dst', feat='iat_size', header=False,
overwrite=False):
"""Load data from data_pth if data_path exists, otherwise, generate data from pcap fiels
Parameters
----------
data_pth:
data_name
direction
feat
header
overwrite
Returns
-------
X: features
y: labels
"""
if overwrite:
if pth.exists(data_pth): os.remove(data_pth)
if not pth.exists(data_pth):
data_pth = generate_data_speed_up(data_name, feat_type=feat, header=header, direction=direction,
out_file=data_pth,
overwrite=overwrite)
return load_data(data_pth)
def _get_data(data_pth=None,
data_name=None,
direction='src_dst',
feat='iat_size',
header=False,
overwrite=False):
if overwrite:
if pth.exists(data_pth): os.remove(data_pth)
if not pth.exists(data_pth):
data_pth = generate_data_speed_up(data_name,
feat_type=feat,
header=header,
direction=direction,
out_file=data_pth,
overwrite=overwrite)
return load_data(data_pth)
def minimal_model_cost(model_name,
model_params_file,
test_file,
params,
project_params_file,
is_parallel=False):
#######################################################################################################
# 1. create a new model from saved parameters
params = {'is_kjl': False, 'is_nystrom': False}
if 'OCSVM' in model_name:
#######################################################################################################
# load params
model_params = load_data(model_params_file)
#######################################################################################################
# create a new model
# 'OCSVM(rbf)':
oc = OCSVM()
oc.kernel = model_params['kernel']
oc._gamma = model_params[
'_gamma'] # only used for 'rbf', 'linear' doeesn't need '_gamma'
oc.gamma = oc._gamma
oc.support_vectors_ = model_params['support_vectors_']
oc._dual_coef_ = model_params[
'_dual_coef_'] # Coefficients of the support vectors in the decision function.
oc.dual_coef_ = oc._dual_coef_
oc._intercept_ = model_params['_intercept_']
oc.intercept_ = oc._intercept_
oc.support_ = np.zeros(
oc.support_vectors_.shape[0], dtype=np.int32
) # np.empty((1,), dtype=np.int32) # # model_params['support_'] # Indices of support vectors.
oc._n_support = model_params[
'_n_support'] # Number of support vectors for each class.
oc._sparse = model_params[
'_sparse'] # spare_kernel_compute or dense_kernel_compute
oc.shape_fit_ = model_params[
'shape_fit_'] # to check if the dimension of train set and test set is the same.
oc.probA_ = np.zeros(
0
) # model_params['probA_'] # /* pairwise probability information */, not used. its values = [].
oc.probB_ = np.zeros(
0
) # model_params['probB_'] # /* pairwise probability information */, not used its values = [].
oc.offset_ = -1 * model_params[
'_intercept_'] # model_params['offset_']
project = None
if 'KJL' in model_name: # KJL-OCSVM
# load params
project_params = load_data(project_params_file)
project = KJL(None)
project.sigma = project_params['sigma']
project.Xrow = project_params['Xrow']
project.U = project_params['U']
params['is_kjl'] = True
elif 'Nystrom' in model_name: # Nystrom-OCSVM
# load params
project_params = load_data(project_params_file)
project = NYSTROM(None)
project.sigma = project_params['sigma']
project.Xrow = project_params['Xrow']
project.eigvec_lambda = project_params['eigvec_lambda']
params['is_nystrom'] = True
elif 'GMM' in model_name:
#######################################################################################################
# load params
model_params = load_data(model_params_file)
# GMM params
oc = GMM()
oc.covariance_type = model_params['covariance_type']
oc.weights_ = model_params['weights_']
oc.means_ = model_params['means_']
# oc.precisions_ = model_params['precisions_']
oc.precisions_cholesky_ = model_params['precisions_cholesky_']
project = None
if 'KJL' in model_name: # KJL-GMM
# load params
project_params = load_data(project_params_file)
project = KJL(None)
project.sigma = project_params['sigma']
project.Xrow = project_params['Xrow']
project.U = project_params['U']
params['is_kjl'] = True
elif 'Nystrom' in model_name: # Nystrom-GMM
# load params
project_params = load_data(project_params_file)
project = NYSTROM(None)
project.sigma = project_params['sigma']
project.Xrow = project_params['Xrow']
project.eigvec_lambda = project_params['eigvec_lambda']
params['is_nystrom'] = True
else:
raise NotImplementedError()
#######################################################################################################
# 2. load test set and evaluate the model
X_test, y_test = load_data(test_file)
# Evaluation
# average time
# minimal model cost
num = 1
if is_parallel:
with Parallel(n_jobs=10,
verbose=0,
backend='loky',
pre_dispatch=1,
batch_size=1) as parallel:
# outs = parallel(delayed(_test)(oc, X_test, y_test, params=params, project=project) for _ in range(num))
outs = parallel(
delayed(_test)(copy.deepcopy(oc),
copy.deepcopy(X_test),
copy.deepcopy(y_test),
params=copy.deepcopy(params),
project=copy.deepcopy(project))
for _ in range(num))
auc, test_time = list(zip(*outs))
auc = np.mean(auc)
test_time = np.mean(test_time)
else:
auc = []
test_time = []
for _ in range(num):
# auc_, test_time_ = _test(oc, X_test, y_test, params, project)
auc_, test_time_ = _test(copy.deepcopy(oc),
copy.deepcopy(X_test),
copy.deepcopy(y_test),
params=copy.deepcopy(params),
project=copy.deepcopy(project))
auc.append(auc_)
test_time.append(test_time_)
auc = np.mean(auc)
test_time = np.mean(test_time)
return auc, X_test, test_time