def getDiademDetectionChildExp(diadem_exp_id, child_type, fold_id, dataset):
def _get_parent_id(diadem_exp_id, child_type):
if dataset is None or dataset == 'all':
return diadem_exp_id
query = session.query(DiademExpAlchemy)
query = query.join(DiademExpAlchemy.exp)
query = query.join(ExpAlchemy.parents)
query = query.filter(ExpAlchemy.kind == 'Detection')
query = query.filter(
ExpRelationshipsAlchemy.parent_id == diadem_exp_id)
query = query.filter(DiademExpAlchemy.type == child_type)
query = query.filter(DiademExpAlchemy.fold_id == fold_id)
return query.one().exp_id
fold_id = None if fold_id == 'None' else int(fold_id)
dataset = None if dataset == 'None' else dataset
if child_type != 'cv':
parent_id = _get_parent_id(diadem_exp_id, child_type)
query = session.query(DiademExpAlchemy)
query = query.join(DiademExpAlchemy.exp)
query = query.join(ExpAlchemy.parents)
query = query.filter(ExpAlchemy.kind == 'Detection')
query = query.filter(ExpRelationshipsAlchemy.parent_id == parent_id)
if dataset is not None and dataset != 'all':
query = query.join(DiademExpAlchemy.dataset)
query = query.filter(DatasetsAlchemy.dataset == dataset)
else:
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == diadem_exp_id)
query = query.filter(DiademExpAlchemy.type == child_type)
query = query.filter(DiademExpAlchemy.fold_id == fold_id)
return jsonify(db_row_to_json(query.one()))
def getAlerts(exp_id, analysis_type):
exp = update_curr_exp(exp_id)
# With proba ? With scores ?
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == exp_id)
diadem_exp = query.one()
with_proba, with_scores = diadem_exp.proba, diadem_exp.with_scoring
# Get alerts
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
if with_proba:
threshold = exp.exp_conf.core_conf.detection_threshold
query = query.filter(PredictionsAlchemy.proba >= threshold)
else:
query = query.filter(PredictionsAlchemy.value == MALICIOUS)
if analysis_type == 'topN' and (with_proba or with_scores):
if with_proba:
query = query.order_by(PredictionsAlchemy.proba.desc())
else:
query = query.order_by(PredictionsAlchemy.score.desc())
elif analysis_type == 'random':
query = call_specific_db_func(secuml_conf.db_type, 'random_order',
(query, ))
query = query.limit(NUM_MAX_DISPLAY)
return jsonify(_predictions_results(query))
def getAlerts(exp_id, analysis_type):
exp = update_curr_exp(exp_id)
# With proba ?
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == exp_id)
with_proba = query.one().proba
threshold = None
if with_proba:
threshold = exp.exp_conf.core_conf.detection_threshold
# Get alerts
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
if with_proba:
query = query.filter(PredictionsAlchemy.proba >= threshold)
if analysis_type == 'topN' and with_proba:
query = query.order_by(PredictionsAlchemy.proba.desc())
elif analysis_type == 'random':
query = call_specific_db_func(secuml_conf.db_type, 'random_order',
(query,))
query = query.limit(TOP_N_ALERTS)
predictions = query.all()
if predictions:
ids, probas = zip(*[(r.instance_id, r.proba) for r in predictions])
else:
ids = []
probas = []
return jsonify({'instances': ids, 'proba': probas})
def get_train_exp(exp_id):
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == exp_id)
row = query.one()
if row.type == 'train':
return exp_id
elif row.type == 'test':
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.diadem_id == row.diadem_id)
query = query.filter(DiademExpAlchemy.fold_id == row.fold_id)
query = query.filter(DiademExpAlchemy.type == 'train')
return query.one().exp_id
else:
assert (False)
def get_feature_user_ids(session, features):
user_ids = [None for _ in range(len(features))]
for i, feature_id in enumerate(features):
query = session.query(FeaturesAlchemy)
query = query.filter(FeaturesAlchemy.id == feature_id)
user_ids[i] = query.one().user_id
return user_ids
def getDatasets(project):
project_id = get_project_id(session, project)
if project_id is None:
return []
query = session.query(DatasetsAlchemy)
query = query.filter(DatasetsAlchemy.project_id == project_id)
return jsonify({'datasets': [r.dataset for r in query.all()]})
def get_train_exp(exp_id):
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == exp_id)
row = query.one()
# get diadem_exp
query = session.query(ExpRelationshipsAlchemy)
query = query.filter(ExpRelationshipsAlchemy.child_id == exp_id)
diadem_exp_id = query.one().parent_id
# get train_exp
query = session.query(DiademExpAlchemy)
query = query.join(DiademExpAlchemy.exp)
query = query.join(ExpAlchemy.parents)
query = query.filter(ExpRelationshipsAlchemy.parent_id == diadem_exp_id)
query = query.filter(ExpAlchemy.kind == 'Train')
query = query.filter(DiademExpAlchemy.fold_id == row.fold_id)
return query.one().exp_id
def getAnnotationsTypes(exp_id, iteration):
query = session.query(IlabExpAlchemy)
query = query.filter(IlabExpAlchemy.id == exp_id)
query = query.filter(IlabExpAlchemy.iter == iteration)
res = query.one()
return jsonify(
{k: getattr(res, k)
for k in ['uncertain', 'malicious', 'benign']})
def getIdent(exp_id, instance_id):
query = session.query(InstancesAlchemy)
query = query.join(InstancesAlchemy.dataset)
query = query.join(DatasetsAlchemy.features)
query = query.join(FeaturesSetsAlchemy.experiments)
query = query.filter(ExpAlchemy.id == exp_id)
query = query.filter(InstancesAlchemy.id == instance_id)
res = query.one()
return jsonify({'ident': res.ident, 'user_id': res.user_instance_id})
def getAlertsClusteringExpId(test_exp_id):
query = session.query(ExpRelationshipsAlchemy)
query = query.join(ExpRelationshipsAlchemy.child)
query = query.join(ExpAlchemy.diadem_exp)
query = query.filter(ExpRelationshipsAlchemy.parent_id == test_exp_id)
query = query.filter(DiademExpAlchemy.type == 'alerts')
try:
return str(query.one().child_id)
except NoResultFound:
return 'None'
def _get_parent_id(diadem_exp_id, child_type):
if dataset is None or dataset == 'all':
return diadem_exp_id
query = session.query(DiademExpAlchemy)
query = query.join(DiademExpAlchemy.exp)
query = query.join(ExpAlchemy.parents)
query = query.filter(ExpAlchemy.kind == 'Detection')
query = query.filter(
ExpRelationshipsAlchemy.parent_id == diadem_exp_id)
query = query.filter(DiademExpAlchemy.type == child_type)
query = query.filter(DiademExpAlchemy.fold_id == fold_id)
return query.one().exp_id
def get_train_exp(exp_id):
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == exp_id)
row = query.one()
if row.type == 'train':
return exp_id
elif row.type == 'test':
# get diadem_exp
query = session.query(ExpRelationshipsAlchemy)
query = query.filter(ExpRelationshipsAlchemy.child_id == exp_id)
diadem_exp_id = query.one().parent_id
# get train_exp
query = session.query(DiademExpAlchemy)
query = query.join(DiademExpAlchemy.exp)
query = query.join(ExpAlchemy.parents)
query = query.filter(
ExpRelationshipsAlchemy.parent_id == diadem_exp_id)
query = query.filter(DiademExpAlchemy.fold_id == row.fold_id)
query = query.filter(DiademExpAlchemy.type == 'train')
return query.one().exp_id
else:
assert (False)
def getDiademChildExp(diadem_exp_id, child_type, fold_id):
fold_id = None if fold_id == 'None' else int(fold_id)
query = session.query(DiademExpAlchemy)
if child_type != 'cv':
query = query.join(DiademExpAlchemy.exp)
query = query.join(ExpAlchemy.parents)
query = query.filter(
ExpRelationshipsAlchemy.parent_id == diadem_exp_id)
else:
query = query.filter(DiademExpAlchemy.exp_id == diadem_exp_id)
query = query.filter(DiademExpAlchemy.type == child_type)
query = query.filter(DiademExpAlchemy.fold_id == fold_id)
return jsonify(db_row_to_json(query.one()))
def getFeaturesInfo(exp_id):
exp = update_curr_exp(exp_id)
features_set_id = exp.exp_conf.features_conf.set_id
query = session.query(FeaturesAlchemy)
query = query.filter(FeaturesAlchemy.set_id == features_set_id)
return jsonify({
res.id: {
'type': res.type,
'user_id': res.user_id,
'name': res.name,
'description': res.description
}
for res in query.all()
})
def getPredictionsScores(exp_id, range_, label):
score_min, score_max = [float(x) for x in range_.split(' - ')]
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
query = query.filter(PredictionsAlchemy.score >= score_min)
query = query.filter(PredictionsAlchemy.score <= score_max)
query = query.order_by(PredictionsAlchemy.score.asc())
if label != 'all':
query = query.join(PredictionsAlchemy.instance)
query = query.join(InstancesAlchemy.ground_truth)
query = query.filter(GroundTruthAlchemy.label == label)
query = call_specific_db_func(secuml_conf.db_type, 'random_order',
(query, ))
query = query.limit(NUM_MAX_DISPLAY)
return jsonify(_predictions_results(query))
def _get_errors(exp_id, fn_fp):
if fn_fp == 'FN':
predicted_value = BENIGN
ground_truth = label_str_to_bool(MALICIOUS)
else:
predicted_value = MALICIOUS
ground_truth = label_str_to_bool(BENIGN)
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
query = query.filter(PredictionsAlchemy.value == predicted_value)
query = query.join(PredictionsAlchemy.instance)
query = query.filter(InstancesAlchemy.label == ground_truth)
query = call_specific_db_func(secuml_conf.db_type, 'random_order',
(query, ))
query = query.limit(NUM_MAX_DISPLAY)
return _predictions_results(query)
def getPredictionsProbas(exp_id, index, label):
index = int(index)
proba_min = index * 0.1
proba_max = (index + 1) * 0.1
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
query = query.filter(PredictionsAlchemy.proba >= proba_min)
query = query.filter(PredictionsAlchemy.proba <= proba_max)
query = query.order_by(PredictionsAlchemy.proba.asc())
if label != 'all':
query = query.join(PredictionsAlchemy.instance)
query = query.join(InstancesAlchemy.ground_truth)
query = query.filter(GroundTruthAlchemy.label == label)
query = call_specific_db_func(secuml_conf.db_type, 'random_order',
(query, ))
query = query.limit(NUM_MAX_DISPLAY)
return jsonify(_predictions_results(query))
def _get_errors(exp_id, fn_fp):
if fn_fp == 'FN':
predicted_value = 'benign'
ground_truth = 'malicious'
else:
predicted_value = 'malicious'
ground_truth = 'benign'
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
query = query.filter(PredictionsAlchemy.value == predicted_value)
query = query.join(PredictionsAlchemy.instance)
query = query.join(InstancesAlchemy.ground_truth)
query = query.filter(GroundTruthAlchemy.label == ground_truth)
query = call_specific_db_func(secuml_conf.db_type, 'random_order',
(query, ))
query = query.limit(NUM_MAX_DISPLAY)
return _predictions_results(query)
def getAllExperiments(project, dataset):
query = session.query(ExpAlchemy)
query = query.join(ExpAlchemy.features_set)
query = query.join(FeaturesSetsAlchemy.dataset)
query = query.outerjoin(ExpAlchemy.parents)
query = query.filter(DatasetsAlchemy.project == project)
query = query.filter(DatasetsAlchemy.dataset == dataset)
query = query.filter(ExpRelationshipsAlchemy.parent_id == null())
experiments = {}
for exp in query.all():
if exp.kind not in experiments:
experiments[exp.kind] = []
experiments[exp.kind].append({'name': exp.name, 'id': exp.id})
for k, v in experiments.items():
t = [(x['id'], x['name']) for x in v]
t.sort(key=operator.itemgetter(0), reverse=True)
experiments[k] = t
return jsonify(experiments)
def getFeaturesInfo(exp_id):
exp = update_curr_exp(exp_id)
features_types_f = path.join(exp.output_dir(), 'features_types.json')
with open(features_types_f, 'r') as f:
features_types = json.load(f)
types_descriptions = {}
for feature_id, feature_type in features_types.items():
query = session.query(FeaturesAlchemy)
query = query.filter(FeaturesAlchemy.id == feature_id)
res = query.one()
user_id, name, description = res.user_id, res.name, res.description
types_descriptions[feature_id] = {
'type': feature_type,
'user_id': user_id,
'name': name,
'description': description
}
return jsonify(types_descriptions)
def getPredictions(exp_id, index, label):
index = int(index)
proba_min = index * 0.1
proba_max = (index + 1) * 0.1
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
query = query.filter(PredictionsAlchemy.proba >= proba_min)
query = query.filter(PredictionsAlchemy.proba <= proba_max)
if label != 'all':
query = query.join(PredictionsAlchemy.instance)
query = query.join(InstancesAlchemy.ground_truth)
query = query.filter(GroundTruthAlchemy.label == label)
predictions = query.all()
if predictions:
ids, probas = zip(*[(r.instance_id, r.proba) for r in predictions])
else:
ids = []
probas = []
return jsonify({'instances': ids, 'proba': probas})
def getDiademChildExp(diadem_exp_id, child_type, fold_id):
fold_id = None if fold_id == 'None' else int(fold_id)
query = session.query(DiademExpAlchemy)
if child_type != 'cv':
query = query.join(DiademExpAlchemy.exp)
query = query.join(ExpAlchemy.parents)
query = query.filter(
ExpRelationshipsAlchemy.parent_id == diadem_exp_id)
else:
query = query.filter(DiademExpAlchemy.exp_id == diadem_exp_id)
query = query.filter(DiademExpAlchemy.type == child_type)
query = query.filter(DiademExpAlchemy.fold_id == fold_id)
row = query.one()
return jsonify({
'exp_id': row.exp_id,
'perf_monitoring': row.perf_monitoring,
'model_interpretation': row.model_interpretation,
'predictions_interpretation': row.predictions_interpretation,
'alerts': row.alerts
})
def getPredictions(exp_id, predicted_value, right_wrong, multiclass):
multiclass = multiclass == 'true'
query = session.query(PredictionsAlchemy)
query = query.filter(PredictionsAlchemy.exp_id == exp_id)
query = query.filter(PredictionsAlchemy.value == predicted_value)
if right_wrong != 'all':
query = query.join(PredictionsAlchemy.instance)
query = query.join(InstancesAlchemy.ground_truth)
field = 'family' if multiclass else 'label'
if right_wrong == 'right':
query = query.filter(
getattr(GroundTruthAlchemy, field) == predicted_value)
elif right_wrong == 'wrong':
query = query.filter(
getattr(GroundTruthAlchemy, field) != predicted_value)
else:
assert (False)
query = call_specific_db_func(secuml_conf.db_type, 'random_order',
(query, ))
query = query.limit(NUM_MAX_DISPLAY)
return jsonify(_predictions_results(query))
def hasGroundTruth(project, dataset):
query = session.query(DatasetsAlchemy)
query = query.filter(DatasetsAlchemy.project == project)
query = query.filter(DatasetsAlchemy.dataset == dataset)
return str(query.one().ground_truth_hash is not None)
def getDatasets(project):
query = session.query(DatasetsAlchemy)
query = query.filter(DatasetsAlchemy.project == project)
return jsonify({'datasets': [r.dataset for r in query.all()]})
def getProjects():
query = session.query(DatasetsAlchemy.project).distinct()
return jsonify({'projects': [r.project for r in query.all()]})
def getRcdClusteringId(exp_id, iteration):
query = session.query(RcdClusteringExpAlchemy)
query = query.filter(RcdClusteringExpAlchemy.id == exp_id)
query = query.filter(RcdClusteringExpAlchemy.iter == iteration)
return jsonify({'clustering_exp_id': query.one().clustering_exp})
def getDiademExp(exp_id):
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == exp_id)
return jsonify(db_row_to_json(query.one()))
def predictionsInterpretation(exp_id):
query = session.query(DiademExpAlchemy)
query = query.filter(DiademExpAlchemy.exp_id == exp_id)
# first() and not one()
# because a train experiment can be shared by several DIADEM experiments.
return str(query.first().predictions_interpretation)
def get_user_instance_ids(instance_ids):
query = session.query(InstancesAlchemy)
query = query.filter(InstancesAlchemy.id.in_(instance_ids))
query = query.order_by(InstancesAlchemy.id)
return [r.user_instance_id for r in query.all()]