trasgodp.numerical package

Module contents

DP mechanisms for numerical columns with clipped results.

trasgodp.numerical.dp_clip_gaussian(df: DataFrame, column: str, epsilon: float, delta=0.001, lower_bound=None, upper_bound=None, new_column=False) → DataFrame

Apply the Gaussian mechanism to a dataframe numeric column and clip the result.

Parameters:

  • df (pandas dataframe) – dataframe with the data under study.

  • columm (string) – column to which the DP mechanism will be applied.

  • epsilon (float) – privacy budget.

  • delta (float) – probability of exceeding the privacy budget.

  • lower_bound (float) – lower bound for clipping and calculating the sensitivity.

  • upper_bound (float) – upper bound for clipping and calculating the sensitivity.

  • new_column (boolean) – boolean, default to False. If False, the new values obtained with the mechanims applied are stored in the same column. If True, a new column ‘dp_{column}’ is created with the new values.

Returns:

dataframe with the column transformed applying the mechanism.

Return type:

pandas dataframe.

trasgodp.numerical.dp_clip_gaussian_array(data: List | ndarray, epsilon: float, delta=0.001, lower_bound=None, upper_bound=None) → ndarray

Apply the Gaussian mechanism to a list or numpy array.

Parameters:

  • data (list or numpy array) – dataset with the data under study.

  • epsilon (float) – privacy budget.

  • delta (float) – probability of exceeding the privacy budget.

  • lower_bound (float) – lower bound for clipping and calculating the sensitivity.

  • upper_bound (float) – upper bound for clipping and calculating the sensitivity.

Returns:

array with the data transformed applying the Gaussian mechanism.

Return type:

numpy array.

trasgodp.numerical.dp_clip_laplace(df: DataFrame, column: str, epsilon: float, lower_bound=None, upper_bound=None, new_column=False) → DataFrame

Apply the Laplace mechanism to a dataframe numeric column and clip the result.

Parameters:

  • df (pandas dataframe) – dataframe with the data under study.

  • columm (string) – column to which the DP mechanism will be applied.

  • epsilon (float) – privacy budget.

  • lower_bound (float) – lower bound for clipping and calculating the sensitivity.

  • upper_bound (float) – upper bound for clipping and calculating the sensitivity.

  • new_column (boolean) – boolean, default to False. If False, the new values obtained with the mechanims applied are stored in the same column. If True, a new column “dp_{column}” is created with the new values.

Returns:

dataframe with the column transformed applying the mechanism.

Return type:

pandas dataframe.

trasgodp.numerical.dp_clip_laplace_array(data: List | ndarray, epsilon: float, lower_bound=None, upper_bound=None) → ndarray

Apply the Laplace mechanism to a list or numpy array.

Parameters:

  • data (list or numpy array) – dataset with the data under study.

  • epsilon (float) – privacy budget.

  • lower_bound (float) – lower bound for clipping and calculating the sensitivity.

  • upper_bound (float) – upper bound for clipping and calculating the sensitivity.

Returns:

array with the data transformed applying the Laplace mechanism.

Return type:

numpy array.