def fromJson(obj, db, cursor):
conf = ProjectionConfFactory.getFactory().fromJson(obj['conf'])
experiment = ProjectionExperiment(obj['project'], obj['dataset'], db,
cursor)
Experiment.expParamFromJson(experiment, obj)
experiment.setConf(conf)
return experiment
def generateParser(parser, binary=True):
Experiment.projectDatasetFeturesParser(parser)
al_group = ActiveLearningConfiguration.generateActiveLearningArguments(
parser)
ActiveLearningConfiguration.generateSupervisedLearningArguments(
parser, binary=binary)
return al_group
def __init__(self, project, dataset, db, cursor, experiment_name=None):
Experiment.__init__(self,
project,
dataset,
db,
cursor,
experiment_name=experiment_name)
self.kind = 'Validation'
def fromJson(obj, db, cursor):
experiment = ClassificationExperiment(obj['project'], obj['dataset'],
db, cursor)
Experiment.expParamFromJson(experiment, obj)
classification_conf = ClassifierConfFactory.getFactory().fromJson(
obj['classification_conf'], experiment)
experiment.setClassifierConf(classification_conf)
return experiment
def initLabels(self, labels_filename=None, overwrite=True):
if isinstance(self.conf, SemiSupervisedProjectionConfiguration):
if labels_filename is None:
message = 'Semi supervised projections require annotated instances. '
message += 'labels_filename must be specified.'
raise ValueError(message)
Experiment.initLabels(self,
labels_filename=labels_filename,
overwrite=overwrite)
def fromJson(obj, db, cursor):
experiment = ActiveLearningExperiment(obj['project'], obj['dataset'], db, cursor)
Experiment.expParamFromJson(experiment, obj)
experiment.labeling_method = obj['labeling_method']
# Validation configuration
experiment.validation_conf = None
if obj['validation_conf'] is not None:
experiment.validation_conf = TestConfiguration.fromJson(obj['validation_conf'], experiment)
experiment.conf = ActiveLearningConfFactory.getFactory().fromJson(obj['conf'], experiment)
return experiment
def loadTrueLabels(self):
labels_file = dir_tools.getDatasetDirectory(self.project, self.dataset)
labels_file += 'labels/true_labels.csv'
# Loads the true labels in the table TrueLabels if the file exists
# Otherwise the table TrueLabels is not created
if not dir_tools.checkFileExists(labels_file):
print >> sys.stderr, 'No ground truth labels for this dataset'
return
exp = Experiment(self.project,
self.dataset,
self.db,
self.cursor,
experiment_name='true_labels')
exp.initLabels('true_labels.csv')
def __init__(self,
project,
dataset,
db,
cursor,
experiment_name=None,
experiment_label=None,
parent=None):
Experiment.__init__(self,
project,
dataset,
db,
cursor,
experiment_name=experiment_name,
experiment_label=experiment_label,
parent=parent)
self.kind = 'Classification'
def generateParser(parser):
Experiment.projectDatasetFeturesParser(parser)
parser.add_argument('--num-folds', type=int, default=4)
parser.add_argument('--multilabel', action='store_true', default=False)
sample_weight_help = 'When set to True, the detection model is learned with '
sample_weight_help += 'sample weights inverse to the proportion of the family '
sample_weight_help += 'in the dataset. Useless if the families are not specified.'
parser.add_argument('--sample-weight',
action='store_true',
default=False,
help=sample_weight_help)
## Validation parameters
validation_help = 'Validation parameters: \n '
validation_help += 'The detection model is validated with a proportion of '
validation_help += 'the instances in the input dataset, or with a separate validation'
validation_help += ' dataset. By default 10% of the instances are used for validation'
validation_group = parser.add_argument_group(validation_help)
validation_group.add_argument('--test-size', type=float, default=0.1)
validation_group.add_argument('--validation-dataset', default=None)
## Alerts
alerts_group = parser.add_argument_group('Alerts parameters')
alerts_group.add_argument(
'--top-n-alerts',
default=100,
help='Number of most confident alerts displayed.')
alerts_group.add_argument(
'--detection-threshold',
type=float,
default=0.8,
help=
'An alert is raised if the predicted probability of maliciousness '
+ 'is above this threshold.')
alerts_group.add_argument(
'--clustering-algo',
default='Kmeans',
choices=['Kmeans', 'GaussianMixture'],
help='Clustering algorithm to analyse the alerts.')
alerts_group.add_argument(
'--num-clusters',
type=int,
default=4,
help='Number of clusters built from the alerts.')
def toJson(self):
conf = Experiment.toJson(self)
conf['__type__'] = 'ActiveLearningExperiment'
conf['labeling_method'] = self.labeling_method
if self.validation_conf is not None:
conf['validation_conf'] = self.validation_conf.toJson()
else:
conf['validation_conf'] = None
conf['conf'] = self.conf.toJson()
return conf
def __init__(self,
project,
dataset,
db,
cursor,
conf,
experiment_name=None,
experiment_label=None,
parent=None):
Experiment.__init__(self,
project,
dataset,
db,
cursor,
experiment_name=experiment_name,
experiment_label=experiment_label,
parent=parent)
self.kind = 'Clustering'
self.conf = conf
def generateParser(parser):
Experiment.projectDatasetFeturesParser(parser)
parser.add_argument(
'--families-supervision',
action='store_true',
default=False,
help=
'When set to True, the semi-supervision is based on the families '
+
'instead of the binary labels. Useless if an unsupervised projection method is used.'
)
parser.add_argument(
'--labels',
'-l',
dest='labels_file',
default=None,
help='CSV file containing the labels of some instances. ' +
'These labels are used for semi-supervised projections.')
parser.add_argument('--num-components', type=int, default=None)
def generateParser(parser):
# Generic arguments
Experiment.projectDatasetFeturesParser(parser)
# Clustering arguments
parser.add_argument('--num-clusters', type=int, default=4)
label_help = 'The clustering is built from all the instances in the dataset, '
label_help += 'or only from the benign or malicious ones. '
label_help += 'By default, the clustering is built from all the instances. '
label_help += 'The malicious and benign instances are selected according to '
label_help += 'the ground truth labels stored in labels/true_labels.csv.'
parser.add_argument('--label',
choices=['all', 'malicious', 'benign'],
default='all',
help=label_help)
# Projection arguments
projection_group = parser.add_argument_group('Projection parameters')
projection_group.add_argument(
'--projection-algo',
choices=['Pca', 'Rca', 'Lda', 'Lmnn', 'Nca', 'Itml', None],
default=None,
help='Projection performed before building the clustering. ' +
'By default the instances are not projected.')
projection_group.add_argument(
'--families-supervision',
action='store_true',
default=False,
help=
'When set to True, the semi-supervision is based on the families '
+
'instead of the binary labels. Useless if an unsupervised projection method is used.'
)
projection_group.add_argument(
'--labels',
'-l',
dest='labels_file',
default=None,
help='CSV file containing the labels of some instances. ' +
'These labels are used for semi-supervised projections.')
def fromJson(obj, db, cursor):
experiment = DescriptiveStatisticsExperiment(obj['project'],
obj['dataset'], db,
cursor)
Experiment.expParamFromJson(experiment, obj)
return experiment
def toJson(self):
conf = Experiment.toJson(self)
conf['__type__'] = 'ClassificationExperiment'
conf['classification_conf'] = self.classification_conf.toJson()
return conf
def __init__(self, project, dataset, db, cursor):
Experiment.__init__(self, project, dataset, db, cursor)
self.kind = 'ActiveLearning'
self.labeling_method = None
def toJson(self):
conf = Experiment.toJson(self)
conf['__type__'] = 'ClusteringExperiment'
conf['conf'] = self.conf.toJson()
return conf
def fromJson(obj, db, cursor):
conf = ClusteringConfFactory.getFactory().fromJson(obj['conf'])
experiment = ClusteringExperiment(obj['project'], obj['dataset'], db,
cursor, conf)
Experiment.expParamFromJson(experiment, obj)
return experiment
def toJson(self):
conf = Experiment.toJson(self)
conf['__type__'] = 'DescriptiveStatisticsExperiment'
return conf
def toJson(self):
conf = Experiment.toJson(self)
conf['__type__'] = 'ProjectionExperiment'
conf['conf'] = self.conf.toJson()
return conf
def fromJson(obj, db, cursor):
experiment = ValidationExperiment(obj['project'], obj['dataset'], db,
cursor)
Experiment.expParamFromJson(experiment, obj)
return experiment
def toJson(self):
conf = Experiment.toJson(self)
conf['__type__'] = 'ValidationExperiment'
return conf
def __init__(self, project, dataset, db, cursor):
Experiment.__init__(self, project, dataset, db, cursor)
self.kind = 'DescriptiveStatistics'