opendp 0.14.2

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

#[cfg(feature = "ffi")]
use polars::prelude::CompatLevel;
#[cfg(feature = "ffi")]
use polars::series::Series;
use polars::{
    error::{PolarsResult, polars_bail},
    prelude::{Column, ColumnsUdf, Expr, GetOutput, lit},
};
#[cfg(feature = "ffi")]
use polars_arrow as arrow;
use polars_plan::prelude::FunctionOptions;
use serde::{Deserialize, Serialize};

use crate::{
    core::{Measurement, MetricSpace},
    domains::{ExprDomain, ExprPlan, WildExprDomain},
    error::Fallible,
    measurements::{PrivateExpr, expr_dp_quantile::DPQuantileShim, expr_noise::NoiseExprMeasure},
    metrics::L01InfDistance,
    polars::{OpenDPPlugin, apply_plugin, match_shim},
    transformations::{StableExpr, traits::UnboundedMetric},
};

#[derive(Clone, Serialize, Deserialize)]
pub(crate) struct DPMedianShim;
impl ColumnsUdf for DPMedianShim {
    // makes it possible to downcast the AnonymousFunction trait object back to Self
    fn as_any(&self) -> &dyn std::any::Any {
        self
    }

    fn call_udf(&self, _: &mut [Column]) -> PolarsResult<Option<Column>> {
        polars_bail!(InvalidOperation: "OpenDP expressions must be passed through make_private_lazyframe to be executed.")
    }
}

impl OpenDPPlugin for DPMedianShim {
    const NAME: &'static str = "dp_median";
    const SHIM: bool = true;
    fn function_options() -> FunctionOptions {
        FunctionOptions::aggregation()
    }

    fn get_output(&self) -> Option<GetOutput> {
        Some(GetOutput::same_type())
    }
}

/// Make a dp median expression measurement.
///
/// # Arguments
/// * `input_domain` - ExprDomain
/// * `input_metric` - The metric space under which neighboring LazyFrames are compared
/// * `expr` - The expression to which the noise will be added
/// * `global_scale` - (Re)scale the noise parameter for the noise distribution
pub fn make_expr_dp_median<MI: 'static + UnboundedMetric, MO: NoiseExprMeasure>(
    input_domain: WildExprDomain,
    input_metric: L01InfDistance<MI>,
    output_measure: MO,
    expr: Expr,
    global_scale: Option<f64>,
) -> Fallible<Measurement<WildExprDomain, L01InfDistance<MI>, MO, ExprPlan>>
where
    Expr: StableExpr<L01InfDistance<MI>, L01InfDistance<MI>> + PrivateExpr<L01InfDistance<MI>, MO>,
    (ExprDomain, MO::Metric): MetricSpace,
{
    let Some([input, candidates, scale]) = match_shim::<DPMedianShim, _>(&expr)? else {
        return fallible!(MakeMeasurement, "Expected {} function", DPMedianShim::NAME);
    };

    let input = apply_plugin(
        vec![input, lit(0.5f64), candidates, scale],
        expr,
        DPQuantileShim,
    );

    input.make_private(input_domain, input_metric, output_measure, global_scale)
}

#[cfg(feature = "ffi")]
#[pyo3_polars::derive::polars_expr(output_type=Null)]
fn dp_median(_: &[Series]) -> PolarsResult<Series> {
    polars_bail!(InvalidOperation: "OpenDP expressions must be passed through make_private_lazyframe to be executed.")
}