def time_feature_extraction(rep=1):
w = Workspace(days=1, empty=True)
set_name = 'mixed_dga_grouped_family_50000_59_0.pkl'
w.load(set_name, settings.SetTypes.mixed_dga_grouped_family.value)
domains = [ld.domain for ld in w.data_sets_loaded['mixed_dga_grouped_family_50000_59_0.pkl'].full]
t = Timer(lambda: feature_extraction.extract_all_features(domains))
print(t.timeit(number=rep))
def train_mixed_dga(clf_types=['svm', 'rf'], n_jobs=-1):
w = Workspace(days=1, empty=True)
w.load_all(settings.SetTypes.mixed_dga_grouped_family.value)
parallel = Parallel(n_jobs=n_jobs, verbose=1)
parallel(
delayed(_training)(clf_type, s)
for s in w.data_sets_loaded.values()
for clf_type in clf_types
)
def start_single_grouped_day_logo(n_jobs=-1, clf_type=None):
w = Workspace(days=1, empty=True)
w.load_all(settings.SetTypes.single_dga_grouped_day.value)
if not clf_type:
eval_train_test.logo_cv('svm', w.data_sets_loaded.values(), n_jobs=n_jobs)
eval_train_test.logo_cv('rf', w.data_sets_loaded.values(), n_jobs=n_jobs)
elif clf_type == 'rf':
eval_train_test.logo_cv('rf', w.data_sets_loaded.values(), n_jobs=n_jobs)
elif clf_type == 'svm':
eval_train_test.logo_cv('svm', w.data_sets_loaded.values(), n_jobs=n_jobs)
def start_mix_dga_kfold(repetitions=5, n_jobs=-1, clf_type=None):
w = Workspace(days=1, empty=True)
w.load_all(settings.SetTypes.mixed_dga_grouped_family.value)
if not clf_type:
eval_train_test.kfold_cv('svm_mix', w.data_sets_loaded.values(), repetitions=repetitions, n_jobs=n_jobs)
eval_train_test.kfold_cv('rf_mix', w.data_sets_loaded.values(), repetitions=repetitions, n_jobs=n_jobs)
elif clf_type == 'rf':
eval_train_test.kfold_cv('rf_mix', w.data_sets_loaded.values(), repetitions=repetitions, n_jobs=n_jobs)
elif clf_type == 'svm':
eval_train_test.kfold_cv('svm_mix', w.data_sets_loaded.values(), repetitions=repetitions, n_jobs=n_jobs)
def predict_exact_or_threshold(clf_type=None, nomix=True, threshold=1, exact=1):
w = Workspace(days=1, empty=True)
w.load_all(settings.SetTypes.mixed_dga_grouped_family.value)
clfs = classifiers.ClassificationEnsemble(nomix=nomix, only_type=clf_type, only_mix=False)
ds = w.data_sets_loaded['mixed_dga_grouped_family_50000_59_2.pkl']
dmns, lbls, groups = ds.expand()
lbls, predicted = clfs.predict_exactly_one_or_threshold(dmns, lbls, threshold=threshold, exact=exact)
stats = Statistic(set_id='mixed_dga_grouped_family_50000_59_2.pkl',
id='threshold_test')
stats.add_run(lbls, predicted, dmns)
return stats
def train_all_available_dga(clf_type=None):
"""
Trains all available single DGAs.
:param clf_type: 'svm', 'rf', None (None means for both)
:return:
"""
w = Workspace(days=1, empty=True)
w.load_all(settings.SetTypes.single_dga.value)
if not clf_type:
for ct in ['svm', 'rf']:
for s in w.data_sets_loaded.values():
_training(ct, s)
elif clf_type == 'rf':
for s in w.data_sets_loaded.values():
_training(clf_type, s)
elif clf_type == 'svm':
for s in w.data_sets_loaded.values():
_training(clf_type, s)
def predict_all_mixed_sets_on_x(n_jobs=8):
w = Workspace(days=1, empty=True)
w.load_all(settings.SetTypes.mixed_dga_grouped_family.value)
clfs = classifiers.ClassificationEnsemble().clfs
clfs = [c for c in clfs if 'mix' not in c.dga_type]
parallel = Parallel(n_jobs=n_jobs, verbose=1)
ds = w.data_sets_loaded['mixed_dga_grouped_family_50000_59_2.pkl']
dmns, lbls, groups = ds.expand()
# res is list of tuples: clf.clf_type, clf.dga_type, lbls, pred_lbl
res = parallel(
delayed(eval_train_test.predict_all_on_x)(clf, ds, dmns, lbls, groups)
for clf in clfs
)
data.serialize_keep_copy(settings.ANALYSIS_FOLDER + '/x_vs_all_results.pkl', (res, dmns))
def start_x_trained_y_test(n_jobs=-1, clf_type=None, x='siemens'):
if x == 'rwth':
load_clfs_from = '/work/ss930620/dga_detection_workspace/clfs/'
elif x == 'siemens':
load_clfs_from = '/work/ss930620/dga_detection_workspace_siemens/clfs/'
w = Workspace(days=1, empty=True)
w.load_all(settings.SetTypes.mixed_dga_grouped_family.value)
test_data_sets = w.data_sets_loaded.values()
if not clf_type:
eval_train_test.trained_clfs_vs_sets(classifiers.ClassificationEnsemble(only_mix=True, clfs_from_path=load_clfs_from).clfs, test_data_sets, n_jobs=n_jobs)
elif clf_type == 'rf':
eval_train_test.trained_clfs_vs_sets(
classifiers.ClassificationEnsemble(only_mix=True, only_type='rf', clfs_from_path=load_clfs_from).clfs, test_data_sets,
n_jobs=n_jobs)
elif clf_type == 'svm':
eval_train_test.trained_clfs_vs_sets(
classifiers.ClassificationEnsemble(only_mix=True, only_type='svm', clfs_from_path=load_clfs_from).clfs,
test_data_sets,
n_jobs=n_jobs)
def time_training(rep=1):
w = Workspace(days=1, empty=True)
set_name = 'mixed_dga_grouped_family_50000_59_0.pkl'
w.load(set_name, settings.SetTypes.mixed_dga_grouped_family.value)
dom, lab, _ = w.data_sets_loaded['mixed_dga_grouped_family_50000_59_0.pkl'].expand()
svm = SVMClassifier(dga='mix')
t = Timer(lambda: svm.training(dom, lab))
print('SVM Training: ' + str(t.timeit(number=rep)))
rf = RFClassifier(dga='mix')
t = Timer(lambda: rf.training(dom, lab))
print('RF Training: ' + str(t.timeit(number=rep)))
set_name = 'mixed_dga_grouped_family_50000_59_1.pkl'
w.load(set_name, settings.SetTypes.mixed_dga_grouped_family.value)
dom, lab, _ = w.data_sets_loaded['mixed_dga_grouped_family_50000_59_1.pkl'].expand()
t = Timer(lambda: svm.predict(dom, lab))
print('SVM Classify: ' + str(t.timeit(number=rep)))
t = Timer(lambda: rf.predict(dom, lab))
print('RF Classify: ' + str(t.timeit(number=rep)))