package differential_privacy

Get desktop application:
View/edit binary Protocol Buffers messages

Used in: BoundedMeanSummary, BoundedVarianceSummary, BoundsSummary

• repeated int64 pos_bin_count = 1

• repeated int64 neg_bin_count = 2

• repeated ValueType input = 2

  Store all inputs.

Next ID: 8

• optional int64 count = 1

  Count of the data subset.

• repeated ValueType pos_sum = 2

  Partial sum data for the dataset.

• repeated ValueType neg_sum = 3

• optional ApproxBoundsSummary bounds_summary = 4

  (DEPRECATED) ApproxBounds data if available. Use `bounds` field instead.

• optional BoundsSummary bounds = 7

  BoundsProvider data if available.

• optional BoundedSumSummary sum_summary = 5

  These two fields are only used by the Java library (not C++), similarly to pos_sum and count.

• optional CountSummary count_summary = 6

• optional BoundedQuantilesSummary.Algorithm algorithm = 11

• optional double epsilon = 2

  Common metadata set for both algorithms.

• optional double lower = 5

• optional double upper = 6

• optional int32 max_partitions_contributed = 7

• optional int32 max_contributions_per_partition = 8

• optional double delta = 3

  Quantile tree specific metadata. Distribution of the data subset, stored in the form of a quantile tree

• map<int32, int64> quantile_tree = 1

• optional MechanismType mechanism_type = 4

• optional int32 tree_height = 9

• optional int32 branching_factor = 10

• repeated double data = 13

  Exponential divide & conquer specific metadata: The list of data point to compute quantiles for.

Used in: BoundedQuantilesSummary

• TREE = 0

• EXPONENTIAL_DIVIDE_AND_CONQUER = 1

Next ID: 13

Used in: BoundedMeanSummary, BoundedVarianceSummary

• repeated ValueType pos_sum = 1

  Partial sum data for the dataset. For automatically set bounds, partial sum values are stored corresponding to each ApproxBounds bin. For manually set bounds, clamped sum will be stored in pos_sum. Currently, used only by C++ library.

• repeated ValueType neg_sum = 2

  neg_sum is used only when bounds are determined automatically.

• optional BoundsSummary bounds = 12

  BoundsProvider data if available.

• optional ValueType partial_sum = 4

  partial_sum is used by the Java library to store partial sum. TODO: Use partial_sum in C++ library when bounds are set manually.

• optional double epsilon = 5

  TODO: Use below fields in C++ library. partial_sum is used by Java library to store sum Sum parameters:

• optional double delta = 6

• optional MechanismType mechanism_type = 7

• optional double lower = 8

• optional double upper = 9

• optional int32 max_partitions_contributed = 10

• optional int32 max_contributions_per_partition = 11

Used for BoundedVariance and BoundedStandardDeviation algorithms. Next ID: 11

• optional int64 count = 1

  Count of the dataset.

• repeated ValueType pos_sum = 2

  Partial sum data for the dataset. For manually set bounds, the clamped sum will be stored in pos_sum. For automatically set bounds, partial sum values stored corresponding to each ApproxBounds bin.

• repeated ValueType neg_sum = 3

• repeated double pos_sum_of_squares = 4

  Partial sum of squares for the dataset. For manually set bounds, clamped sum of squares is stored in pos_sum_of_squares.

• repeated double neg_sum_of_squares = 5

• optional ApproxBoundsSummary bounds_summary = 6

  (DEPRECATED) ApproxBounds data if available. Use `bounds` field instead.

• optional BoundsSummary bounds = 10

  BoundsProvider data if available.

• optional BoundedSumSummary sum_of_squares_summary = 7

  These three fields are only used by the Java library (not C++), similarly to pos_sum_of_squares, pos_sum and count.

• optional BoundedSumSummary sum_summary = 8

• optional CountSummary count_summary = 9

Accuracy information about results of automatic bounding algorithms. When ApproxBounds is called by bounded algorithms, the BoundingReport can be used to pass differentially private intermediate results to help users understand the accuracy implications of the output.

Used in: Output.ErrorReport

• optional ValueType lower_bound = 1

  Lower and upper bounds produced by the ApproxBounds algorithm.

• optional ValueType upper_bound = 2

• optional double num_inputs = 3

  Noisy number of total inputs to the bounding algorithm.

• optional double num_outside = 4

  Noisy number of inputs lying outside the bounds.

Next ID: 3

Used in: BoundedMeanSummary, BoundedSumSummary, BoundedVarianceSummary

• oneof bounds_summary

  • ApproxBoundsSummary approx_bounds_summary = 1

Used in: Output.Element

• double upper_bound = 1

• double lower_bound = 2

• double confidence_level = 3

  The percentile confidence level. For 95% CI, this value is 0.95.

Used in: BoundedMeanSummary, BoundedVarianceSummary

• optional int64 count = 1

  Count of the data subset.

• optional double epsilon = 3

  TODO: Use below fields in C++ library. Count parameters:

• optional double delta = 4

• optional MechanismType mechanism_type = 5

• optional int32 max_partitions_contributed = 6

• optional int32 max_contributions_per_partition = 7

• repeated string element = 1

• repeated int64 bin_count = 1

• optional ValueType partial_sum = 1

• optional double epsilon = 2

• optional double delta = 3

• optional MechanismType mechanism_type = 4

• optional int64 lower = 5

• optional int64 upper = 6

• optional int32 max_partitions_contributed = 7

• optional int32 max_contributions_per_partition = 8

Used in: BoundedQuantilesSummary, BoundedSumSummary, CountSummary, LongBoundedSumSummary

• EMPTY = 0

• LAPLACE = 1

• GAUSSIAN = 2

• DISCRETE_LAPLACE = 3

• MECHANISM_NONE = 5

Output data produced by a differentially private algorithm.

• repeated Output.Element elements = 1

• optional Output.ErrorReport error_report = 3

  Error report is attached if either the noise confidence interval or the bounding report is available.

Used in: Output

• optional ValueType value = 1

  Diff. priv. result of the operation performed over the input data.

• optional ConfidenceInterval noise_confidence_interval = 3

  Confidence interval of the noise added to the result

Contains information about algorithm accuracy.

Used in: Output

• optional BoundingReport bounding_report = 2

• optional int64 ids_count = 1

  The count of unique privacy units IDs in the partition.

• optional double epsilon = 2

  PreAggSelectPartition parameters:

• optional double delta = 3

• optional int32 max_partitions_contributed = 4

• optional int32 pre_threshold = 5

Serialized summary data of a subset of the input data, to be merged at a later time.

• optional google.protobuf.Any data = 2

  The summary data.

Defining our own value type to restrict the acceptable data types. It would change as per the future extensions.

Used in: BinarySearchSummary, BoundedMeanSummary, BoundedSumSummary, BoundedVarianceSummary, BoundingReport, LongBoundedSumSummary, Output.Element

• oneof value

  • int64 int_value = 1

  • double float_value = 2

  • string string_value = 3