composition

Measurement for combining multiple measurements into a single measurement.

Classes

Composition

Describes a measurement constructed by composing two or more Measurements.

class Composition(measurements, hint=None)

Bases: tmlt.core.measurements.base.Measurement

Describes a measurement constructed by composing two or more Measurements.

Parameters:

• measurements (Sequence[tmlt.core.measurements.base.Measurement]) –

• hint (Optional[Callable[[Any, Any], Tuple[Any, Ellipsis]]]) –

property measurements: List[tmlt.core.measurements.base.Measurement]

Returns the list of measurements being composed.

Return type:

List[tmlt.core.measurements.base.Measurement]

property input_domain: tmlt.core.domains.base.Domain

Return input domain for the measurement.

Return type:

tmlt.core.domains.base.Domain

property input_metric: tmlt.core.metrics.Metric

Distance metric on input domain.

Return type:

tmlt.core.metrics.Metric

property output_measure: tmlt.core.measures.Measure

Distance measure on output.

Return type:

tmlt.core.measures.Measure

property is_interactive: bool

Returns true iff the measurement is interactive.

Return type:

bool

__init__(measurements, hint=None)

Constructor.

It supports PureDP, ApproxDP, and RhoZCDP. Input metrics, domains, and output measures must be identical across all supplied measurements.

Parameters:

• measurements (Sequence[Measurement]) – List of measurements to be composed. The provided measurements must all have PureDP, all have RhoZCDP, or all have ApproxDP as their output_measure.

• hint (Optional[Callable[[Any, Any], Tuple[Any, ...]]]) – An optional hint. A hint is only required if one or more of the measurements’ privacy_function()’s raise NotImplementedError. The hint takes in the same arguments as privacy_relation(), and should return a d_out for each measurement to be composed, where all of the d_outs sum to less than the d_out passed into the hint.

privacy_function(d_in)

Returns the smallest d_out satisfied by the measurement.

Returns the sum of the privacy_function()’s of the composed measurements on d_in (adding element-wise for ApproxDP).

Parameters:

d_in (Any) – Distance between inputs under input_metric.

Raises:

NotImplementedError – If the privacy_function() of one of the composed measurements raises NotImplementedError.

Return type:

Any

privacy_relation(d_in, d_out)

Returns True only if outputs are close under close inputs.

Let d_outs be the d_out from the privacy_function()’s of all measurements or the d_outs from the hint if one of them raises NotImplementedError.

And total_d_out to be the sum of d_outs (adding element-wise for ApproxDP ).

This returns True if total_d_out <= d_out (the input argument) and each composed measurement satisfies its privacy_relation() from d_in to its d_out from d_outs.

Parameters:

• d_in (Any) – Distance between inputs under input_metric.

• d_out (Any) – Distance between outputs under output_measure.

Raises:

ValueError – If a hint is not provided and the privacy_function() of one of the composed measurements raises NotImplementedError.

Return type:

bool

__call__(data)

Return answers to composed measurements.

Parameters:

data (Any) –

Return type:

List