opendp 0.4.0

A library of differential privacy algorithms for the statistical analysis of sensitive private data.
Documentation 
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
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276

use std::any::Any;
use std::rc::Rc;

use crate::core::{Domain, Function, Measure, Measurement, Metric, PrivacyRelation};
use crate::dom::AllDomain;
use crate::error::*;
use crate::traits::{CheckNull, InfSub};

/// A structure tracking the state of an interactive measurement queryable.
/// It's generic over state (S), query (Q), answer (A), so it can be used for any
/// interactive measurement expressible as a transition function.
pub struct Queryable<S, Q, A> {
    /// The state of the Queryable. It is wrapped in an option so that ownership can be moved out
    /// temporarily, during transitions.
    state: Option<S>,
    /// The transition function of the Queryable. Takes the current state and a query, returns
    /// the new state and the answer.
    transition: Rc<dyn Fn(S, &Q) -> Fallible<(S, A)>>,
}

impl<S, Q, A> Queryable<S, Q, A> {
    /// Constructs a Queryable with initial state and transition function.
    pub fn new(initial: S, transition: impl Fn(S, &Q) -> Fallible<(S, A)> + 'static) -> Self {
        Queryable {
            state: Some(initial),
            transition: Rc::new(transition),
        }
    }

    /// Evaluates a query.
    pub fn eval(&mut self, query: &Q) -> Fallible<A> {
        // Take temporary ownership of the state from this struct.
        let state = self.state.take().unwrap_assert("Queryable state is only accessed in this method, always replaced.");
        // Obtain then new state and answer.
        let new_state_answer = (self.transition)(state, query)?;
        // Restore ownership of the state into this struct.
        self.state.replace(new_state_answer.0);
        Ok(new_state_answer.1)
    }
}

impl<S, Q> Queryable<S, Q, Box<dyn Any>> {
    /// Evaluates a polymorphic query and downcasts to the given type.
    pub fn eval_poly<A: 'static>(&mut self, query: &Q) -> Fallible<A> {
        self.eval(query)?.downcast().map_err(|_| err!(FailedCast)).map(|b| *b)
    }
}

impl<S, Q, A> CheckNull for Queryable<S, Q, A> { fn is_null(&self) -> bool { false } }

pub type InteractiveMeasurement<DI, DO, MI, MO, S, Q> = Measurement<DI, AllDomain<Queryable<S, Q, <DO as Domain>::Carrier>>, MI, MO>;

/// The state of an adaptive composition Queryable.
pub struct AcState<DI: Domain, DO: Domain, MI: Metric, MO: Measure> {
    input_domain: DI,
    output_domain: DO,
    input_metric: MI,
    output_measure: MO,
    d_in_budget: MI::Distance,
    d_out_budget: MO::Distance,
    data: DI::Carrier,
}
impl<DI: Domain, DO: Domain, MI: Metric, MO: Measure> AcState<DI, DO, MI, MO> where MO::Distance: PartialOrd + InfSub {
    pub fn new(
        input_domain: DI,
        output_domain: DO,
        input_metric: MI,
        output_measure: MO,
        data: DI::Carrier,
        d_in_budget: MI::Distance,
        d_out_budget: MO::Distance,
    ) -> Self {
        Self {
            input_domain,
            output_domain,
            input_metric,
            output_measure,
            data,
            d_in_budget,
            d_out_budget,
        }
    }

    /// Checks that a measurement (of a query) is compatible with this Queryable state.
    fn check_types(&self, measurement: &Measurement<DI, DO, MI, MO>) -> Fallible<()> {
        if measurement.input_domain != self.input_domain {
            return fallible!(DomainMismatch, "wrong query input domain")
        } else if measurement.output_domain != self.output_domain {
            return fallible!(DomainMismatch, "wrong query output domain")
        } else if measurement.input_metric != self.input_metric {
            return fallible!(MetricMismatch, "wrong query input metric")
        } else if measurement.output_measure != self.output_measure {
            return fallible!(MeasureMismatch, "wrong query output measure")
        }
        Ok(())
    }

    /// Checks that there is adequate budget in this Queryable state.
    fn check_budget(&self, privacy_relation: &PrivacyRelation<MI, MO>, d_out_query: &MO::Distance) -> Fallible<()> {
        privacy_relation.eval(&self.d_in_budget, d_out_query)?;
        if d_out_query > &self.d_out_budget {
            return fallible!(FailedRelation, "not enough budget")
        }
        Ok(())
    }

    /// Updates this Queryable state by consuming the given amount of budget.
    fn update(self, d_out_query: &MO::Distance) -> Fallible<Self> {
        Ok(Self { d_out_budget: self.d_out_budget.neg_inf_sub(d_out_query)?, ..self })
    }

    /// Processes a query, generating a new Queryable state.
    fn transition(self, (measurement, d_out_query): &AcQuery<DI, DO, MI, MO>) -> Fallible<(Self, DO::Carrier)>
        where MO::Distance: Clone + PartialOrd + InfSub {
        self.check_types(measurement)?;
        self.check_budget(&measurement.privacy_relation, d_out_query)?;
        let res = measurement.invoke(&self.data)?;
        let new = self.update(d_out_query)?;
        Ok((new, res))
    }
}
type AcQuery<DI, DO, MI, MO> = (Measurement<DI, DO, MI, MO>, <MO as Measure>::Distance);
type AcQueryable<DI, DO, MI, MO> = Queryable<AcState<DI, DO, MI, MO>, AcQuery<DI, DO, MI, MO>, <DO as Domain>::Carrier>;
type AcMeasurement<DI, DO, MI, MO> = InteractiveMeasurement<DI, DO, MI, MO, AcState<DI, DO, MI, MO>, AcQuery<DI, DO, MI, MO>>;

pub fn make_adaptive_composition<DI, DO, MI, MO>(
    input_domain: DI,
    output_domain: DO,
    input_metric: MI,
    output_measure: MO,
    d_in_budget: MI::Distance,
    d_out_budget: MO::Distance,
) -> AcMeasurement<DI, DO, MI, MO>
    where DI: 'static + Domain,
          DI::Carrier: Clone,
          DO: 'static + Domain,
          MI: 'static + Metric,
          MI::Distance: 'static + Clone + PartialOrd,
          MO: 'static + Measure,
          MO::Distance: 'static + Clone + PartialOrd + InfSub {
    AcMeasurement::new(
        input_domain.clone(),
        AllDomain::new(),
        Function::new(enclose!((input_domain, input_metric, output_measure, d_in_budget, d_out_budget), move |arg: &DI::Carrier| -> AcQueryable<DI, DO, MI, MO> {
            AcQueryable::new(
                // TODO: Remove these clones and have the Queryable use refs? (Also remove Clone trait bounds.)
                AcState::new(input_domain.clone(), output_domain.clone(), input_metric.clone(), output_measure.clone(), arg.clone(), d_in_budget.clone(), d_out_budget.clone()),
                |s, q| s.transition(q))
        })),
        input_metric,
        output_measure,
        PrivacyRelation::new(move |d_in, d_out| d_in <= &d_in_budget && d_out <= &d_out_budget),
    )
}


#[cfg(test)]
mod tests {
    use crate::dist::{MaxDivergence, AbsoluteDistance, SymmetricDistance};
    use crate::dom::VectorDomain;
    use crate::error::*;
    use crate::meas::*;
    use crate::poly::PolyDomain;
    use crate::trans::*;

    use super::*;

    fn make_dummy_meas<TO: From<i32> + CheckNull>() -> Measurement<AllDomain<i32>, AllDomain<TO>, AbsoluteDistance<f64>, MaxDivergence<f64>> {
        Measurement::new(
            AllDomain::new(),
            AllDomain::new(),
            Function::new(|a: &i32| TO::from(a.clone())),
            AbsoluteDistance::<f64>::default(),
            MaxDivergence::<f64>::default(),
            PrivacyRelation::new(|d_in, d_out| d_out <= d_in),
        )
    }

    #[test]
    fn test_adaptive_comp_simple() -> Fallible<()> {
        let meas1 = make_dummy_meas::<i32>();
        let meas2 = make_dummy_meas::<i32>();

        let data = 999;
        let d_in_budget = 1.0;
        let d_out_budget = 1.0;
        let adaptive = make_adaptive_composition(meas1.input_domain.clone(), meas1.output_domain.clone(), meas1.input_metric.clone(), meas1.output_measure.clone(), d_in_budget, d_out_budget);
        let mut queryable = adaptive.invoke(&data)?;
        let res1 = queryable.eval(&(meas1, d_out_budget / 2.0))?;
        assert_eq!(res1, 999);
        let res2 = queryable.eval(&(meas2, d_out_budget / 2.0))?;
        assert_eq!(res2, 999);

        Ok(())
    }

    #[test]
    fn test_adaptive_composition_poly() -> Fallible<()> {
        let meas1 = make_dummy_meas::<i32>().into_poly();
        let meas2 = make_dummy_meas::<i64>().into_poly();

        let data = 999;
        let d_in_budget = 1.0;
        let d_out_budget = 1.0;
        let adaptive = make_adaptive_composition(meas1.input_domain.clone(), PolyDomain::new(), meas1.input_metric.clone(), meas1.output_measure.clone(), d_in_budget, d_out_budget);
        let mut queryable = adaptive.invoke(&data)?;
        let res1: i32 = queryable.eval_poly(&(meas1, d_out_budget / 2.0))?;
        assert_eq!(res1, 999_i32);
        let res2: i64 = queryable.eval_poly(&(meas2, d_out_budget / 2.0))?;
        assert_eq!(res2, 999_i64);

        Ok(())
    }

    #[test]
    fn test_adaptive_composition_no_budget() -> Fallible<()> {
        let meas1 = make_dummy_meas::<i32>();
        let meas2 = make_dummy_meas::<i32>();

        let data = 999;
        let d_in_budget = 1.0;
        let d_out_budget = 1.0;
        let adaptive = make_adaptive_composition(meas1.input_domain.clone(), meas1.output_domain.clone(), meas1.input_metric.clone(), meas1.output_measure.clone(), d_in_budget, d_out_budget);
        let mut queryable = adaptive.invoke(&data)?;
        let res1 = queryable.eval(&(meas1, d_out_budget / 2.0))?;
        assert_eq!(res1, 999);
        let res2 = queryable.eval(&(meas2, d_out_budget));
        // TODO: Would be handy to have a way of comparing Errors for this assertion.
        assert!(res2.is_err());

        Ok(())
    }

    #[test]
    fn test_adaptive_composition_chain() -> Fallible<()> {
        // Definitions
        let input_domain = VectorDomain::new(AllDomain::new());
        let output_domain = PolyDomain::new();
        let input_metric = SymmetricDistance::default();
        let output_measure = MaxDivergence::default();

        // Build queryable
        let d_in = 1;
        let d_out_budget = 1.0;
        let adaptive = make_adaptive_composition(input_domain, output_domain, input_metric, output_measure, d_in, d_out_budget);
        let data = vec![0.6, 2.8, 6.0, 9.4, 8.9, 7.7, 5.9, 3.4, 8.0, 2.4, 4.4, 7.1, 6.0, 3.2, 7.1];
        let mut queryable = adaptive.invoke(&data)?;
        // NO FURTHER ACCESS TO DATA AFTER THIS POINT.

        // Set parameters for queries
        let count_bounds = (0, 20);
        let val_bounds = (0.0, 10.0);
        let d_out_query = 0.5 * d_out_budget;

        // Noisy count
        let measurement1 = (
            make_count()? >>
            make_base_geometric(1.0 / d_out_query, Some(count_bounds))?
        )?.into_poly();
        let query1 = (measurement1, d_out_query);
        let _result1: i32 = queryable.eval_poly(&query1)?;
        // println!("_result = {}", result1);

        // Noisy sum
        let measurement2 = (
            make_clamp(val_bounds)? >>
            make_bounded_sum(val_bounds)? >>
            make_base_laplace(1.0 / d_out_query)?
        )?.into_poly();
        let query2 = (measurement2, d_out_query);
        let _result2: f64 = queryable.eval_poly(&query2)?;
        // println!("_result = {}", result2);

        Ok(())
    }
}