mihirkhandekar

PostalCode,Latitude,Longitude
118547,1.2762,103.7931
118552,1.2765,103.7924
118546,1.2761,103.7931
118549,1.2769,103.7932
118555,1.2773,103.7908
118553,1.2766,103.792
118551,1.2764,103.7928
118554,1.277,103.7909
118556,1.2775,103.7897

mihirkhandekar

dataset_path = "datasets/purchase100.txt"

datahandler = advreg_wrapper.get_datahandler(dataset_path=dataset_path, 
											 batch_size=128, 
											 training_size=20000)

mihirkhandekar

def classification_nn_model(input_features):
	initializer = tf.compat.v1.keras.initializers.random_normal(0.0, 0.01)
	model = tf.keras.Sequential(
		  [
		 	keraslayers.Dense(
		 		512, 
		 		activation = tf.nn.tanh, 
		 		input_shape = (input_features,), 
		 		kernel_initializer = initializer,
		 		bias_initializer = 'zeros'

mihirkhandekar

# `saved_path` is required for obtaining the training data that was used to
# train the target classification model, in "npy" format
saved_path = "datasets/purchase100.txt.npy"

# Similar to `saved_path` being used to form the memberset for attack model, 
# `dataset_path` is used for forming the non-member set of the training data of
# attack model.
dataset_path = 'datasets/purchase100.txt'

datahandlerA = ml_privacy_meter.utils.attack_data.attack_data(dataset_path=dataset_path,

mihirkhandekar

# Load the classification model by passing the model, that
# you'd like to train, in `model` argument
advreg_wrapper.load_classification(model=cmodel, 
                                   optimizer="adam", 
                                   learning_rate=0.001)

advreg_wrapper.load_inference(optimizer="adam", learning_rate=0.0001)

# begins training in an adversarial manner (with defense if _lambda > 0)
advreg_wrapper.begin_advreg_training(datahandler=datahandler, 

mihirkhandekar

attackobj = ml_privacy_meter.attack.meminf.initialize(
				target_train_model=cmodelA, 
				target_attack_model=cmodelA,
				train_datahandler=datahandlerA,
				attack_datahandler=datahandlerA,
				layers_to_exploit=[3, 4],
				gradients_to_exploit=[4])
attackobj.train_attack()

mihirkhandekar

def classification_nn_model(input_features):
	initializer = tf.compat.v1.keras.initializers.random_normal(0.0, 0.01)
	model = tf.keras.Sequential(
		  [
		 	keraslayers.Dense(
		 		512, 
		 		activation = tf.nn.tanh, 
		 		input_shape = (input_features,), 
		 		kernel_initializer = initializer,
		 		bias_initializer = 'zeros'