def fromJson(obj, secuml_conf):
conf = ClassifierConfFactory.getFactory().fromJson(
obj['classification_conf'])
experiment = ClassificationExperiment(secuml_conf)
experiment.initExperiment(obj['project'], obj['dataset'], create=False)
Experiment.expParamFromJson(experiment, obj, conf)
return experiment
def fromJson(obj):
classification_factory = ClassifierConfFactory.getFactory()
classification_conf = classification_factory.fromJson(
obj['classification_conf'])
conf = RareCategoryDetectionStrategy(classification_conf,
obj['cluster_strategy'],
obj['num_annotations'],
obj['cluster_weights'])
return conf
def fromJson(obj):
validation_conf = None
if obj['validation_conf'] is not None:
validation_conf = ValidationDatasetConf.fromJson(
obj['validation_conf'])
binary_model_conf = ClassifierConfFactory.getFactory().fromJson(
obj['models_conf']['binary'])
conf = RandomSamplingConfiguration(obj['auto'], obj['budget'], obj['batch'],
binary_model_conf, validation_conf)
return conf
def gornitzBinaryModelConf(logger):
classifier_args = {}
classifier_args['num_folds'] = 4
classifier_args['sample_weight'] = False
classifier_args['families_supervision'] = False
test_conf = UnlabeledLabeledConf()
classifier_args['test_conf'] = test_conf
binary_model_conf = ClassifierConfFactory.getFactory().fromParam(
'Sssvdd', classifier_args, logger=logger)
return binary_model_conf
def fromJson(obj):
validation_conf = None
if obj['validation_conf'] is not None:
validation_conf = ValidationDatasetConf.fromJson(
obj['validation_conf'])
binary_model_conf = ClassifierConfFactory.getFactory().fromJson(
obj['models_conf']['binary'])
conf = AladinConfiguration(obj['auto'], obj['budget'],
obj['num_annotations'], binary_model_conf,
validation_conf)
return conf
def setExperimentFromArgs(self, args):
factory = ClassifierConfFactory.getFactory()
if args.model == 'AlreadyTrained':
self.already_trained = args.model_exp_id
conf = self.generateAlreadyTrainedConf(factory, args, self.logger)
else:
conf = factory.fromArgs(args.model, args, logger=self.logger)
self.setConf(conf,
args.features_file,
annotations_filename='ground_truth.csv')
self.export()
def generateParamsFromArgs(args):
supervised_args = {}
supervised_args['num_folds'] = 4
supervised_args['sample_weight'] = False
supervised_args['families_supervision'] = False
test_conf = UnlabeledLabeledConf()
supervised_args['test_conf'] = test_conf
binary_model_conf = ClassifierConfFactory.getFactory().fromParam(
'Sssvdd', supervised_args)
params = ActiveLearningConfiguration.generateParamsFromArgs(
args, binary_model_conf=binary_model_conf)
params['batch'] = args.batch
return params
def getMulticlassModel(self):
params = {}
params['num_folds'] = 4
params['sample_weight'] = False
params['families_supervision'] = True
params['optim_algo'] = 'liblinear'
params['alerts_conf'] = None
test_conf = UnlabeledLabeledConf(logger=self.alerts_conf.logger)
params['test_conf'] = test_conf
conf = ClassifierConfFactory.getFactory().fromParam(
'LogisticRegression', params, logger=self.alerts_conf.logger)
model = conf.model_class(conf)
return model
def fromJson(obj):
validation_conf = None
if obj['validation_conf'] is not None:
validation_conf = ValidationDatasetConf.fromJson(
obj['validation_conf'])
multiclass_model_conf = ClassifierConfFactory.getFactory().fromJson(
obj['models_conf']['multiclass'])
rare_category_detection_conf = RareCategoryDetectionStrategy.fromJson(
obj['rare_category_detection_conf'])
conf = RareCategoryDetectionConfiguration(
obj['auto'], obj['budget'], rare_category_detection_conf,
multiclass_model_conf, validation_conf)
return conf
def generateParamsFromArgs(args, logger=None):
supervised_args = {}
supervised_args['num_folds'] = 4
supervised_args['sample_weight'] = False
supervised_args['families_supervision'] = False
test_conf = UnlabeledLabeledConf(logger=logger)
supervised_args['test_conf'] = test_conf
binary_model_conf = ClassifierConfFactory.getFactory().fromParam(
'LogisticRegression', supervised_args, logger=logger)
params = ActiveLearningConfiguration.generateParamsFromArgs(
args, binary_model_conf=binary_model_conf, logger=logger)
params['num_annotations'] = args.num_annotations
return params
def aladinMulticlassModelConf(logger):
classifier_args = {}
classifier_args['num_folds'] = 4
classifier_args['sample_weight'] = False
classifier_args['families_supervision'] = True
classifier_args['alerts_conf'] = None
classifier_args['optim_algo'] = 'liblinear'
test_conf = UnlabeledLabeledConf(logger=logger)
classifier_args['test_conf'] = test_conf
factory = ClassifierConfFactory.getFactory()
multiclass_model_conf = factory.fromParam('LogisticRegression',
classifier_args,
logger=logger)
return multiclass_model_conf
def fromJson(obj):
validation_conf = None
if obj['validation_conf'] is not None:
validation_conf = ValidationDatasetConf.fromJson(
obj['validation_conf'])
rare_category_detection_conf = RareCategoryDetectionStrategy.fromJson(
obj['rare_category_detection_conf'])
binary_model_conf = ClassifierConfFactory.getFactory().fromJson(
obj['models_conf']['binary'])
conf = IlabConfiguration(obj['auto'], obj['budget'],
rare_category_detection_conf,
obj['num_uncertain'], obj['eps'],
binary_model_conf,
validation_conf)
return conf
def generateParamsFromArgs(args):
params = ActiveLearningConfiguration.generateParamsFromArgs(args)
multiclass_classifier_args = {}
multiclass_classifier_args['num_folds'] = args.num_folds
multiclass_classifier_args['sample_weight'] = False
multiclass_classifier_args['families_supervision'] = True
multiclass_classifier_args['alerts_conf'] = None
test_conf = UnlabeledLabeledConf()
multiclass_classifier_args['test_conf'] = test_conf
multiclass_conf = ClassifierConfFactory.getFactory().fromParam(
args.model_class, multiclass_classifier_args)
rare_category_detection_conf = RareCategoryDetectionStrategy(multiclass_conf,
args.cluster_strategy, args.num_annotations, 'uniform')
params['rare_category_detection_conf'] = rare_category_detection_conf
params['num_annotations'] = args.num_annotations
params['multiclass_model_conf'] = multiclass_conf
return params
def generateParamsFromArgs(args):
params = ActiveLearningConfiguration.generateParamsFromArgs(args)
multiclass_classifier_args = {}
multiclass_classifier_args['num_folds'] = args.num_folds
multiclass_classifier_args['sample_weight'] = False
multiclass_classifier_args['families_supervision'] = True
multiclass_classifier_args['optim_algo'] = 'liblinear'
test_conf = UnlabeledLabeledConf()
multiclass_classifier_args['test_conf'] = test_conf
multiclass_conf = ClassifierConfFactory.getFactory().fromParam(
'LogisticRegression', multiclass_classifier_args)
rare_category_detection_conf = RareCategoryDetectionStrategy(multiclass_conf,
args.cluster_strategy, args.num_annotations, 'uniform')
params['rare_category_detection_conf'] = rare_category_detection_conf
params['num_uncertain'] = args.num_uncertain
params['eps'] = 0.49
return params
def generateParser():
parser = argparse.ArgumentParser(
description='Learn a detection model. ' +
'The ground-truth must be stored in annotations/ground_truth.csv.')
Experiment.projectDatasetFeturesParser(parser)
models = [
'LogisticRegression', 'Svc', 'GaussianNaiveBayes', 'DecisionTree',
'RandomForest', 'GradientBoosting'
]
subparsers = parser.add_subparsers(dest='model')
factory = ClassifierConfFactory.getFactory()
for model in models:
model_parser = subparsers.add_parser(model)
factory.generateParser(model, model_parser)
## Add subparser for already trained model
already_trained = subparsers.add_parser('AlreadyTrained')
factory.generateParser('AlreadyTrained', already_trained)
return parser
def generateParamsFromArgs(args, binary_model_conf=None):
if binary_model_conf is None:
supervised_args = {}
supervised_args['num_folds'] = args.num_folds
supervised_args['sample_weight'] = args.sample_weight
supervised_args['families_supervision'] = False
supervised_args['test_conf'] = UnlabeledLabeledConf()
binary_model_conf = ClassifierConfFactory.getFactory().fromParam(
args.model_class, supervised_args)
active_learning_params = {}
active_learning_params['auto'] = args.auto
active_learning_params['budget'] = args.budget
active_learning_params['binary_model_conf'] = binary_model_conf
validation_conf = None
if args.validation_dataset is not None:
validation_conf = ValidationDatasetConf(args.validation_dataset)
active_learning_params['validation_conf'] = validation_conf
return active_learning_params