1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220

use indexmap::map::IndexMap;
use ndarray::prelude::*;

use crate::{base, Float, proto, Warnable};
use crate::base::{
    AggregatorProperties, DataType, IndexKey,
    NodeProperties, SensitivitySpace, Value, ValueProperties,
    ArrayProperties, Nature, NatureContinuous, Vector1DNull
};
use crate::components::{Component, Expandable, Sensitivity};
use crate::errors::*;
use crate::utilities::prepend;

impl Component for proto::Quantile {
    fn propagate_property(
        &self,
        _privacy_definition: &Option<proto::PrivacyDefinition>,
        public_arguments: IndexMap<base::IndexKey, &Value>,
        mut properties: base::NodeProperties,
        node_id: u32
    ) -> Result<Warnable<ValueProperties>> {
        let mut data_property: ArrayProperties = properties.get::<IndexKey>(&"data".into())
            .ok_or("data: missing")?.array()
            .map_err(prepend("data:"))?.clone();

        if !data_property.releasable {
            data_property.assert_is_not_aggregated()?;
        }

        if data_property.data_type != DataType::Float && data_property.data_type != DataType::Int {
            return Err("data: atomic type must be numeric".into());
        }

        let has_bounds = public_arguments.contains_key(&IndexKey::from("lower"))
            && public_arguments.contains_key(&IndexKey::from("upper"));

        Ok(match properties.remove::<IndexKey>(&"candidates".into()) {
            Some(candidates_property) => {
                let candidates_property: &ArrayProperties = candidates_property.array()
                    .map_err(prepend("candidates:"))?;

                if data_property.data_type != candidates_property.data_type {
                    return Err("data_type of data must match data_type of candidates".into())
                }

                if data_property.num_columns()? != candidates_property.num_columns()? {
                    return Err("candidates is not column-conformable with the data".into())
                }

                // upper bound for n * max(a, 1 - a) - |(1 - a) * #z - a * (n - #z)|
                //               = n * max(a, 1 - a) - |#z - an|
                //              <= n * max(a, 1 - a) (because |#z - an| minimized when #z = an)
                let utility_upper_bound = candidates_property.num_records
                    .map(|n| n as f64 * self.alpha.max(1. - self.alpha));

                // bounds make the quantile nan-robust
                if !has_bounds {
                    data_property.assert_non_null()?;
                    data_property.assert_is_not_empty()?;
                }

                ValueProperties::Array(ArrayProperties {
                    num_records: candidates_property.num_records,
                    num_columns: data_property.num_columns,
                    nullity: candidates_property.nullity,
                    releasable: data_property.releasable && candidates_property.releasable,
                    c_stability: data_property.c_stability.clone(),
                    aggregator: Some(AggregatorProperties::new(
                        proto::component::Variant::Quantile(self.clone()),
                        properties, data_property.num_columns()?)),
                    nature: Some(Nature::Continuous(NatureContinuous {
                        lower: Vector1DNull::Float((0..data_property.num_columns()?)
                            .map(|_| Some(0.)).collect()),
                        upper: Vector1DNull::Float((0..data_property.num_columns()?)
                            .map(|_| utility_upper_bound).collect())
                    })),
                    data_type: DataType::Float,
                    dataset_id: None,
                    node_id: node_id as i64,
                    is_not_empty: true,
                    dimensionality: candidates_property.dimensionality,
                    group_id: data_property.group_id,
                    naturally_ordered: data_property.naturally_ordered,
                    sample_proportion: None
                }).into()
            },
            None => {
                if has_bounds { return Err("bounds are only useful when evaluating candidates".into()) }

                data_property.assert_is_not_empty()?;
                // save a snapshot of the state when aggregating
                data_property.aggregator = Some(AggregatorProperties::new(
                    proto::component::Variant::Quantile(self.clone()),
                    properties,
                    data_property.num_columns()?));

                data_property.num_records = Some(1);

                ValueProperties::Array(data_property).into()
            }
        })
    }
}

impl Sensitivity for proto::Quantile {
    fn compute_sensitivity(
        &self,
        privacy_definition: &proto::PrivacyDefinition,
        properties: &NodeProperties,
        sensitivity_type: &SensitivitySpace,
    ) -> Result<Value> {
        let data_property = properties.get::<IndexKey>(&"data".into())
            .ok_or("data: missing")?.array()
            .map_err(prepend("data:"))?.clone();

        data_property.assert_is_not_aggregated()?;

        match sensitivity_type {
            SensitivitySpace::KNorm(_k) => {
                let lower = data_property.lower_float()?;
                let upper = data_property.upper_float()?;

                let row_sensitivity = lower.iter()
                    .zip(upper.iter())
                    .map(|(min, max)| max - min)
                    .collect::<Vec<Float>>();

                let mut array_sensitivity = Array::from(row_sensitivity).into_dyn();
                array_sensitivity.insert_axis_inplace(Axis(0));

                Ok(array_sensitivity.into())
            }

            // SensitivitySpace::Exponential(implementation) if implementation == "standard" => {
            SensitivitySpace::Exponential => {
                data_property.assert_non_null()?;

                let neighboring_type = Neighboring::from_i32(privacy_definition.neighboring)
                    .ok_or_else(|| Error::from("neighboring definition must be either \"AddRemove\" or \"Substitute\""))?;
                use proto::privacy_definition::Neighboring;
                let cell_sensitivity = match neighboring_type {
                    Neighboring::AddRemove => self.alpha.max(1. - self.alpha),
                    Neighboring::Substitute => 1.
                } as Float;

                let row_sensitivity = (0..data_property.num_columns()?)
                    .map(|_| cell_sensitivity)
                    .collect::<Vec<Float>>();

                let array_sensitivity = Array::from(row_sensitivity).into_dyn();
                // array_sensitivity.insert_axis_inplace(Axis(0));

                Ok(array_sensitivity.into())
            }

            // this implementation is robust to nans, but only supports alpha == .5
            // SensitivitySpace::Exponential(implementation) if implementation == "NaN-robust" => {
            //
            //     if self.alpha != 0.5 {
            //         return Err(Error::from("alpha must be 0.5 to use the NaN-robust quantile"))
            //     }
            //
            //     let neighboring_type = Neighboring::from_i32(privacy_definition.neighboring)
            //         .ok_or_else(|| Error::from("neighboring definition must be either \"AddRemove\" or \"Substitute\""))?;
            //     use proto::privacy_definition::Neighboring;
            //     let cell_sensitivity = match neighboring_type {
            //         Neighboring::AddRemove => self.alpha.max(1. - self.alpha),
            //         Neighboring::Substitute => 1.
            //     } as Float;
            //
            //     let row_sensitivity = (0..data_property.num_columns()?)
            //         .map(|_| cell_sensitivity)
            //         .collect::<Vec<Float>>();
            //
            //     let array_sensitivity = Array::from(row_sensitivity).into_dyn();
            //     // array_sensitivity.insert_axis_inplace(Axis(0));
            //
            //     Ok(array_sensitivity.into())
            // }
            _ => Err("Quantile sensitivity is not implemented for the specified sensitivity space".into())
        }
    }
}


macro_rules! make_quantile {
    ($variant:ident, $alpha:expr, $interpolation:expr) => {

        impl Expandable for proto::$variant {
            fn expand_component(
                &self,
                _privacy_definition: &Option<proto::PrivacyDefinition>,
                component: &proto::Component,
                _public_arguments: &IndexMap<IndexKey, &Value>,
                _properties: &base::NodeProperties,
                component_id: u32,
                _maximum_id: u32,
            ) -> Result<base::ComponentExpansion> {
                let mut expansion = base::ComponentExpansion::default();

                expansion.computation_graph.insert(component_id, proto::Component {
                    arguments: component.arguments.clone(),
                    variant: Some(proto::component::Variant::Quantile(proto::Quantile {
                        alpha: $alpha,
                        interpolation: $interpolation
                    })),
                    omit: component.omit,
                    submission: component.submission,
                });
                expansion.traversal.push(component_id);

                Ok(expansion)
            }
        }
    }
}

make_quantile!(Minimum, 0.0, "lower".to_string());
make_quantile!(Median, 0.5, "midpoint".to_string());
make_quantile!(Maximum, 1.0, "upper".to_string());