opendp.meas module

opendp.meas.make_base_discrete_gaussian(scale, D='AllDomain<int>', MO='ZeroConcentratedDivergence<Q>')

Make a Measurement that adds noise from the discrete_gaussian(scale) distribution to the input. Adjust D to noise vector-valued data.

Parameters:

• scale – noise scale parameter for the distribution. scale == standard_deviation.

• D (Type Arguments) – Domain of the data type to be privatized. Valid values are VectorDomain<AllDomain<T>> or AllDomain<T>

• MO (Type Arguments) – Output measure. The only valid measure is ZeroConcentratedDivergence<Q>, but Q can be f32 or f64

Returns:

A base_discrete_gaussian step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_base_discrete_laplace(scale, D='AllDomain<int>', QO=None)

Make a Measurement that adds noise from the discrete_laplace(scale) distribution to the input. Adjust D to noise vector-valued data. This uses make_base_discrete_laplace_cks20 if scale is greater than 10, otherwise it uses make_base_discrete_laplace_linear.

Parameters:

• scale – noise scale parameter for the distribution. scale == sqrt(2) * standard_deviation.

• D (Type Arguments) – Domain of the data type to be privatized. Valid values are VectorDomain<AllDomain<T>> or AllDomain<T>

• QO (Type Arguments) – Data type of the sensitivity, scale, and budget.

Returns:

A base_discrete_laplace step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_base_discrete_laplace_cks20(scale, D='AllDomain<int>', QO=None)

Make a Measurement that adds noise from the discrete_laplace(scale) distribution to the input. Uses the sampling algorithm from CKS20, The Discrete Gaussian for Differential Privacy. Adjust D to noise vector-valued data.

Parameters:

• scale – noise scale parameter for the distribution. scale == sqrt(2) * standard_deviation.

• D (Type Arguments) – Domain of the data type to be privatized. Valid values are VectorDomain<AllDomain<T>> or AllDomain<T>

• QO (Type Arguments) – Data type of the sensitivity, scale, and budget.

Returns:

A base_discrete_laplace_cks20 step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_base_discrete_laplace_linear(scale, bounds=None, D='AllDomain<int>', QO=None)

Make a Measurement that adds noise from the discrete_laplace(scale) distribution to the input. This algorithm can be executed in constant time if bounds are passed. Adjust D to noise vector-valued data.

Parameters:

• scale – noise scale parameter for the distribution. scale == sqrt(2) * standard_deviation.

• bounds (Any) – Set bounds on the count to make the algorithm run in constant-time.

• D (Type Arguments) – Domain of the data type to be privatized. Valid values are VectorDomain<AllDomain<T>> or AllDomain<T>

• QO (Type Arguments) – Data type of the sensitivity, scale, and budget.

Returns:

A base_discrete_laplace_linear step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_base_gaussian(scale, k=-1074, D='AllDomain<T>', MO='ZeroConcentratedDivergence<T>')

Make a Measurement that adds noise from the gaussian(scale) distribution to the input. Adjust D to noise vector-valued data. The output epsilon may be no greater than one.

Parameters:

• scale – noise scale parameter for the gaussian distribution. scale == standard_deviation.

• k (int) – The noise granularity in terms of 2^k. Larger values are more computationally efficient, but have a looser privacy map. Defaults to the smallest granularity.

• D (Type Arguments) – Domain of the data type to be privatized. Valid values are VectorDomain<AllDomain<T>> or AllDomain<T>

• MO (Type Arguments) – Output measure. The only valid measure is ZeroConcentratedDivergence<T>.

Returns:

A base_gaussian step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_base_geometric(scale, bounds=None, D='AllDomain<int>', QO=None)

Deprecated. Use make_base_discrete_laplace instead (more efficient). make_base_discrete_laplace_linear has a similar interface with the optional constant-time bounds.

Parameters:

• scale – noise scale parameter for the geometric distribution. scale == sqrt(2) * standard_deviation.

• bounds (Any) – Set bounds on the count to make the algorithm run in constant-time.

• D (Type Arguments) – Domain of the data type to be privatized. Valid values are VectorDomain<AllDomain<T>> or AllDomain<T>

• QO (Type Arguments) – Data type of the sensitivity, scale, and budget.

Returns:

A base_geometric step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_base_laplace(scale, k=-1074, D='AllDomain<T>')

Make a Measurement that adds noise from the laplace(scale) distribution to a scalar value. Adjust D to noise vector-valued data.

Parameters:

• scale – Noise scale parameter for the laplace distribution. scale == sqrt(2) * standard_deviation.

• k (int) – The noise granularity in terms of 2^k. Larger values are more computationally efficient, but have a looser privacy map. Defaults to the smallest granularity.

• D (Type Arguments) – Domain of the data type to be privatized. Valid values are VectorDomain<AllDomain<T>> or AllDomain<T>

Returns:

A base_laplace step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_base_ptr(scale, threshold, TK, k=-1074, TV=None)

Make a Measurement that uses propose-test-release to privatize a hashmap of counts.

Parameters:

• scale – Noise scale parameter for the laplace distribution. scale == sqrt(2) * standard_deviation.

• threshold – Exclude counts that are less than this minimum value.

• k (int) – The noise granularity in terms of 2^k. Larger values are more computationally efficient, but have a looser privacy map. Defaults to the smallest granularity.

• TK (Type Arguments) – Type of Key. Must be hashable/categorical.

• TV (Type Arguments) – Type of Value. Must be float.

Returns:

A base_ptr step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_randomized_response(categories, prob, constant_time=False, T=None, Q=None)

Make a Measurement that implements randomized response on a categorical value.

Parameters:

• categories (Any) – Set of valid outcomes

• prob – Probability of returning the correct answer. Must be in [1/num_categories, 1)

• constant_time (bool) – Set to true to enable constant time

• T (Type Arguments) – Data type of a category.

• Q (Type Arguments) – Data type of probability and budget.

Returns:

A randomized_response step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library

opendp.meas.make_randomized_response_bool(prob, constant_time=False, Q=None)

Make a Measurement that implements randomized response on a boolean value.

Parameters:

• prob – Probability of returning the correct answer. Must be in [0.5, 1)

• constant_time (bool) – Set to true to enable constant time

• Q (Type Arguments) – Data type of probability and budget.

Returns:

A randomized_response_bool step.

Return type:

Measurement

Raises:

• AssertionError – if an argument’s type differs from the expected type

• UnknownTypeError – if a type-argument fails to parse

• OpenDPException – packaged error from the core OpenDP library