protected_change

Types for programmatically specifying what changes in input tables are protected.

Classes

ProtectedChange	A description of the largest change in a dataset that is protected under DP.
AddMaxRows	Protect the addition or removal of any set of max_rows rows.
AddOneRow	A shorthand for the common case of AddMaxRows with max_rows = 1.
AddMaxRowsInMaxGroups	Protect the addition or removal of rows across a finite number of groups.
AddRowsWithID	Protect the addition or removal of rows with a specific identifier.

class ProtectedChange

Bases: abc.ABC

A description of the largest change in a dataset that is protected under DP.

A ProtectedChange describes, for a particular table, the largest change that can be made to that table while still being indistinguishable under Tumult Analytics’ DP guarantee. The appropriate protected change to use is one corresponding to the largest possible change to the table when adding or removing a unit of protection, e.g. a person. For more information, see the privacy promise topic guide.

class AddMaxRows

Bases: ProtectedChange

Protect the addition or removal of any set of max_rows rows.

This ProtectedChange is a generalization of the standard “add/remove one row” DP guarantee, hiding the addition or removal of any set of at most max_rows rows from a table.

max_rows :int

The maximum number of rows that may be added or removed.

class AddOneRow

Bases: AddMaxRows

A shorthand for the common case of AddMaxRows with max_rows = 1.

class AddMaxRowsInMaxGroups

Bases: ProtectedChange

Protect the addition or removal of rows across a finite number of groups.

AddMaxRowsInMaxGroups provides a similar guarantee to AddMaxRows, but it uses some additional information to apply less noise in some cases. That information is about groups: collections of rows which share the same value in a particular column. That column would typically be some kind of categorical value, for example a state where a person lives or has lived. Instead of specifying a maximum total number of rows that may be added or removed, AddMaxRowsInMaxGroups limits the number of rows that may be added or removed in any particular group, as well as the maximum total number of groups that may be affected. If these limits are meant to correspond to the maximum contribution of a specific entity to the dataset, that must be enforced before the data is passed to Tumult Analytics.

AddMaxRowsInMaxGroups is intended for advanced use cases, and its use should be considered carefully. Note that it only provides improved accuracy when used with zCDP – with pure DP, it is equivalent to using AddMaxRows with the same total number of rows to be added/removed.

The most common case where AddMaxRowsInMaxGroups is useful is for dealing with datasets that have already undergone some type of preprocessing before being turned over to an analyst. Where possible, it is preferred to do such processing inside of Tumult Analytics instead, as it allows specifying a simpler protected change (e.g. AddRowsWithID) and relying on Analytics’ privacy tracking to handle the complex parts of the analysis.

grouping_column :str

The name of the column specifying the group.

max_groups :int

The maximum number of groups that may differ.

max_rows_per_group :int

The maximum number of rows which may be added to or removed from each group.

class AddRowsWithID

Bases: ProtectedChange

Protect the addition or removal of rows with a specific identifier.

Instead of limiting the number of rows that may be added or removed, AddRowsWithID hides the addition or removal of all rows with the same value in the specified column.

The ID column must be a string, integer (or long), or date; it cannot be a float or a timestamp.

id_column :str

The name of the column containing the identifier.

id_space :str = default_id_space

The identifier space of the rows that may be added or removed. If not specified, a default will be assigned when using this protected change with Session.from_dataframe().