def __init__(self, model=None, training_instances=None):
learner.__init__(self)
if model:
self.set_model(model)
else:
self.model = None
self.training_instances = training_instances
def __init__(self, base_model=None, training_instances=None, attacker=None,
params: Dict = None):
learner.__init__(self)
self.model = Model(base_model)
# self.attack_alg = None # Type: class
# self.adv_params = None
self.attacker = attacker # Type: Adversary
self.set_training_instances(training_instances)
self.iteration_times = 5 # int: control the number of rounds directly
def __init__(self, params=None, training_instances=None):
learner.__init__(self)
self.weight_vector = None
self.bias = 0
self.c_delta = 0.5
if params is not None:
self.set_params(params)
if training_instances is not None:
self.set_training_instances(training_instances)
def __init__(self, training_instances=None, params=None):
learner.__init__(self)
self.weight_vector = None # type: np.array(shape=(1))
self.num_features = 0 # type: int
self.hinge_loss_multiplier = 0.5 # type: float
self.max_feature_deletion = 30 # type: int
self.bias = 0 # type: int
if params is not None:
self.set_params(params)
if training_instances is not None:
self.set_training_instances(training_instances)
def __init__(self, training_instances: List[Instance], n: int, lda=0.1,
verbose=False):
"""
:param training_instances: the instances on which to train
:param n: the number of unpoisoned instances in training_instances - the
size of the original dataset
:param lda: lambda - for regularization term
:param verbose: if True, the solver will be in verbose mode
"""
learner.__init__(self)
self.training_instances = training_instances
self.n = n
self.lda = lda # lambda
self.verbose = verbose
self.num_features = self.training_instances[0].get_feature_count()
self.w = None
self.b = None
def __init__(self):
learner.__init__(self)
raise NotImplementedError