def var(self, column: str, privacy_budget: PrivacyBudget) -> float:
"""Compute the variance.
:param column: Name of the selected column
:param privacy_budget: Privacy budget to be used
:return: Private variance of the selected column
"""
assert column in self._columns, f'Column `{column}`is not exists.'
assert column in self._data_domains
domain = self._data_domains[column]
assert isinstance(domain, RealDataDomain)
var = np.var(self._dataframe[column])
sensitivity = domain.length()**2/len(self._dataframe[column]) # (H-L)^2/N
noisy_var = laplace_mechanism(var, sensitivity, privacy_budget)
self.privacy_budget_tracker.update_privacy_loss(privacy_budget)
return noisy_var
def median(self, column: str, privacy_budget: PrivacyBudget) -> float:
"""Compute the median.
:param column: Name of the selected column
:param privacy_budget: Privacy budget to be used
:return: Private median of the selected column
"""
assert column in self._columns, f'Column `{column}`is not exists.'
assert column in self._data_domains
domain = self._data_domains[column]
assert isinstance(domain, RealDataDomain)
median = np.median(self._dataframe[column])
sensitivity = domain.length()/2
noisy_median = laplace_mechanism(median, sensitivity, privacy_budget)
self.privacy_budget_tracker.update_privacy_loss(privacy_budget)
return noisy_median
def std(self, column: str, privacy_budget: PrivacyBudget) -> float:
"""Compute the standard deviation.
:param column: Name of the selected column
:param privacy_budget: Privacy budget to be used
:return: Private standard deviation of the selected column
"""
assert column in self._columns, f'Column `{column}`is not exists.'
assert column in self._data_domains
domain = self._data_domains[column]
assert isinstance(domain, RealDataDomain)
std = np.std(self._dataframe[column])
sensitivity = domain.length()/np.sqrt(len(self._dataframe[column]))
noisy_std = laplace_mechanism(std, sensitivity, privacy_budget)
self.privacy_budget_tracker.update_privacy_loss(privacy_budget)
return noisy_std