opendp 0.14.2-dev.20260401.2

A library of differential privacy algorithms for the statistical analysis of sensitive private data.
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

use super::*;
use polars::prelude::*;

use crate::{
    domains::{AtomDomain, LazyFrameDomain, Margin, SeriesDomain},
    measurements::{PrivateExpr, make_private_expr, make_private_lazyframe},
    metrics::{FrameDistance, InsertDeleteDistance, L0PInfDistance, SymmetricDistance},
    polars::PrivacyNamespace,
    transformations::test_helper::get_test_data,
};

#[test]
fn test_make_expr_puredp() -> Fallible<()> {
    let (lf_domain, lf) = get_test_data()?;
    let scale: f64 = 0.0;

    let m_quant = make_private_expr(
        lf_domain.select(),
        L0PInfDistance(InsertDeleteDistance),
        MaxDivergence,
        col("const_1f64").dp().sum((lit(0.), lit(1.)), Some(scale)),
        None,
    )?;

    let dp_expr = m_quant.invoke(&lf.logical_plan)?.expr;
    let df_actual = lf.select([dp_expr]).collect()?;

    assert_eq!(df_actual, df!("const_1f64" => [1000.0])?);

    Ok(())
}

#[test]
fn test_make_expr_zcdp() -> Fallible<()> {
    let (lf_domain, lf) = get_test_data()?;
    let scale: f64 = 0.0;

    let m_quant = make_private_expr(
        lf_domain.select(),
        L0PInfDistance(InsertDeleteDistance),
        ZeroConcentratedDivergence::default(),
        col("const_1f64").dp().sum((lit(0.), lit(1.)), Some(scale)),
        None,
    )?;

    let dp_expr = m_quant.invoke(&lf.logical_plan)?.expr;
    let df_actual = lf.select([dp_expr]).collect()?;

    assert_eq!(df_actual, df!("const_1f64" => [1000.0])?);

    Ok(())
}

#[test]
fn test_make_expr_gaussian() -> Fallible<()> {
    let (lf_domain, lf) = get_test_data()?;
    let scale: f64 = 0.0;

    let m_quant = make_private_expr(
        lf_domain.select(),
        L0PInfDistance(InsertDeleteDistance),
        ZeroConcentratedDivergence::default(),
        col("const_1f64").dp().sum((lit(0.), lit(1.)), Some(scale)),
        None,
    )?;

    let dp_expr = m_quant.invoke(&lf.logical_plan)?.expr;
    let df_actual = lf.select([dp_expr]).collect()?;

    assert_eq!(df_actual, df!("const_1f64" => [1000.0])?);

    Ok(())
}

#[test]
fn test_make_laplace_grouped() -> Fallible<()> {
    let (lf_domain, lf) = get_test_data()?;
    let scale: f64 = 0.0;

    let expr_exp = col("chunk_(..10u32)")
        .clip(lit(0), lit(8))
        .sum()
        .dp()
        .noise(None);
    let m_lap = make_private_lazyframe(
        lf_domain,
        FrameDistance(SymmetricDistance),
        MaxDivergence,
        lf.clone().group_by(["chunk_2_bool"]).agg([expr_exp]),
        Some(scale),
        None,
    )?;
    // sum([0, 1, 2, 3, 4]) * 100 = 1000
    // sum([5, 6, 7, 8, 8]) * 100 = 3400

    let df_act = m_lap.invoke(&lf)?.collect()?;
    let df_exp = df!(
        "chunk_2_bool" => [false, true],
        "chunk_(..10u32)" => [1000, 3400]
    )?;

    assert_eq!(
        df_act.sort(["chunk_2_bool"], Default::default())?,
        df_exp.sort(["chunk_2_bool"], Default::default())?
    );
    Ok(())
}

fn check_autocalibration(
    margin: Margin,
    bounds: (u32, u32),
    d_in: (u32, u32, u32),
) -> Fallible<()> {
    let series_domain = SeriesDomain::new("A", AtomDomain::<i32>::default());
    let lf_domain = LazyFrameDomain::new(vec![series_domain])?.with_margin(margin)?;
    let expr_domain = lf_domain.select();

    // Get resulting sum (expression result)
    let m_sum = col("A")
        .clip(lit(bounds.0), lit(bounds.1))
        .sum()
        .dp()
        .noise(None)
        .make_private(
            expr_domain,
            L0PInfDistance(InsertDeleteDistance),
            MaxDivergence,
            Some(1.),
        )?;

    let epsilon = m_sum.map(&d_in)?;
    // autocalibration chooses a noise scale based on the sensitivity.
    // because of this, epsilon will always work out to 1.
    println!("epsilon: {:?}", epsilon);
    assert_eq!(epsilon, 1.0);

    Ok(())
}

#[test]
fn test_sum_unbounded_dp_autocalibration() -> Fallible<()> {
    check_autocalibration(Margin::select().with_max_length(100), (4, 7), (1, 1, 1))
}

#[test]
fn test_sum_bounded_dp_autocalibration() -> Fallible<()> {
    check_autocalibration(
        Margin::select()
            .with_max_length(100)
            .with_invariant_lengths(),
        (4, 7),
        (1, 2, 2),
    )
}