def describe_dataset_in_correlated_attribute_mode(
self,
dataset_file,
k=0,
epsilon=0.1,
attribute_to_datatype_dict={},
categorical_attributes={},
seed=0):
"""Generate dataset description using correlated attribute mode.
Users only need to call this function. It packages the rest functions.
Args:
dataset_file: string, directory and file name of the sensitive dataset as input in csv format.
epsilon: float, a parameter in differential privacy.
attribute_to_datatype_dict: Dict, mappings of {column_name: data_type}, e.g., {"gender": "string"}.
categorical_attributes: Set or List, e.g., {"gender", "nationality"}
seed: int or float, seeding the randomness.
"""
self.describe_dataset_in_independent_attribute_mode(
dataset_file, epsilon, attribute_to_datatype_dict,
categorical_attributes, seed)
self.encoded_dataset = self.encode_dataset_into_interval_indices()
self.bayesian_network = greedy_bayes(
self.input_dataset[self.encoded_dataset.columns], k, epsilon)
self.dataset_description['bayesian_network'] = self.bayesian_network
self.dataset_description[
'conditional_probabilities'] = construct_noisy_conditional_distributions(
self.bayesian_network, self.encoded_dataset, epsilon)
def describe_dataset_in_correlated_attribute_mode(self,
dataset_file,
k=0,
epsilon=0.1,
attribute_to_datatype: Dict[str, DataType] = None,
attribute_to_is_categorical: Dict[str, bool] = None,
attribute_to_is_candidate_key: Dict[str, bool] = None,
categorical_attribute_domain_file: str = None,
numerical_attribute_ranges: Dict[str, List] = None,
seed=0,
bayesian_network: List[tuple] = None,
user_pool: int = 1):
"""Generate dataset description using correlated attribute mode.
Parameters
----------
dataset_file : str
File name (with directory) of the sensitive dataset as input in csv format.
k : int
Maximum number of parents in Bayesian network.
epsilon : float
A parameter in Differential Privacy. Increase epsilon value to reduce the injected noises. Set epsilon=0 to turn
off Differential Privacy.
attribute_to_datatype : dict
Dictionary of {attribute: datatype}, e.g., {"age": "Integer", "gender": "String"}.
attribute_to_is_categorical : dict
Dictionary of {attribute: boolean}, e.g., {"gender":True, "age":False}.
attribute_to_is_candidate_key: dict
Dictionary of {attribute: boolean}, e.g., {"id":True, "name":False}.
categorical_attribute_domain_file: str
File name of a JSON file of some categorical attribute domains.
numerical_attribute_ranges: dict
Dictionary of {attribute: [min, max]}, e.g., {"age": [25, 65]}
seed : int or float
Seed the random number generator.
"""
self.describe_dataset_in_independent_attribute_mode(dataset_file,
epsilon,
attribute_to_datatype,
attribute_to_is_categorical,
attribute_to_is_candidate_key,
categorical_attribute_domain_file,
numerical_attribute_ranges,
seed)
self.df_encoded = self.encode_dataset_into_binning_indices()
if self.df_encoded.shape[1] < 2:
raise Exception("Correlated Attribute Mode requires at least 2 attributes(i.e., columns) in dataset.")
if len(bayesian_network) == 0:
self.bayesian_network = greedy_bayes(self.df_encoded, k, epsilon / 2, user_pool)
else:
print('Pre-configured Bayesian Network (BN) is used')
bn_list = []
for var, parents in bayesian_network.items():
bn_list.append((var, parents))
self.bayesian_network = bn_list
print(self.bayesian_network)
self.data_description['bayesian_network'] = self.bayesian_network
self.data_description['conditional_probabilities'] = construct_noisy_conditional_distributions(
self.bayesian_network, self.df_encoded, epsilon / 2)
def describe_dataset_in_correlated_attribute_mode(
self,
dataset_file,
k=0,
epsilon=0.1,
attribute_to_datatype={},
attribute_to_is_categorical={},
attribute_to_is_candidate_key={},
seed=0):
"""Generate dataset description using correlated attribute mode.
Users only need to call this function. It packages the rest functions.
Parameters
----------
dataset_file : str
File name (with directory) of the sensitive dataset as input in csv format.
k : int
Maximum number of parents in Bayesian network.
epsilon : float
A parameter in differential privacy.
attribute_to_datatype : dict
Mappings of {attribute: datatype}, e.g., {"age": "Integer", "gender": "String"}.
attribute_to_is_categorical : dict
Mappings of {attribute: boolean}, e.g., {"gender":True, "age":False}.
attribute_to_is_candidate_key: dict
Mappings of {attribute: boolean}, e.g., {"id":True, "name":False}.
seed : int or float
Seed the random number generator.
"""
self.describe_dataset_in_independent_attribute_mode(
dataset_file, epsilon, attribute_to_datatype,
attribute_to_is_categorical, attribute_to_is_candidate_key, seed)
self.encoded_dataset = self.encode_dataset_into_binning_indices()
if self.encoded_dataset.shape[1] < 2:
raise Exception(
"Constructing Bayesian Network needs more attributes.")
self.bayesian_network = greedy_bayes(self.encoded_dataset, k, epsilon)
self.dataset_description['bayesian_network'] = self.bayesian_network
self.dataset_description[
'conditional_probabilities'] = construct_noisy_conditional_distributions(
self.bayesian_network, self.encoded_dataset, epsilon)
def describe_dataset_in_correlated_attribute_mode(
self,
dataset_file,
k=0,
epsilon=0.1,
attribute_to_datatype={},
attribute_to_is_categorical={},
seed=0):
"""Generate dataset description using correlated attribute mode.
Users only need to call this function. It packages the rest functions.
Parameters
----------
dataset_file : str
File name (with directory) of the sensitive dataset as input in csv format.
k : int
Maximum number of parents in Bayesian network.
epsilon : float
A parameter in differential privacy.
attribute_to_datatype : dict
Mappings of {attribute: datatype}, e.g., {"age": "int"}.
attribute_to_is_categorical : dict
Mappings of {attribute: boolean},, e.g., {"age":False}.
seed : int or float
Seed the random number generator.
"""
self.describe_dataset_in_independent_attribute_mode(
dataset_file, epsilon, attribute_to_datatype,
attribute_to_is_categorical, seed)
self.encoded_dataset = self.encode_dataset_into_interval_indices()
if self.encoded_dataset.columns.size < 2:
raise Exception(
"Current dataset doesn't have enough attributes to build Bayesian network in correlated attribute mode."
)
self.bayesian_network = greedy_bayes(
self.input_dataset[self.encoded_dataset.columns], k, epsilon)
self.dataset_description['bayesian_network'] = self.bayesian_network
self.dataset_description[
'conditional_probabilities'] = construct_noisy_conditional_distributions(
self.bayesian_network, self.encoded_dataset, epsilon)