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

use std::sync::Arc;

#[cfg(feature = "ffi")]
use polars::series::Series;
use polars::{
    error::{PolarsResult, polars_bail},
    prelude::{AnonymousColumnsUdf, Column, ColumnsUdf, DataType, Expr, Field, len},
};
use polars_plan::prelude::FunctionOptions;
use serde::{Deserialize, Serialize};

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

#[derive(Clone, Serialize, Deserialize)]
pub(crate) struct DPFrameLenShim;

impl ColumnsUdf for DPFrameLenShim {
    fn as_any(&self) -> &dyn std::any::Any {
        self
    }
    fn call_udf(&self, _: &mut [Column]) -> PolarsResult<Column> {
        polars_bail!(InvalidOperation: "OpenDP expressions must be passed through make_private_lazyframe to be executed.")
    }
}

impl AnonymousColumnsUdf for DPFrameLenShim {
    fn as_column_udf(self: std::sync::Arc<Self>) -> std::sync::Arc<dyn ColumnsUdf> {
        self
    }

    fn deep_clone(self: std::sync::Arc<Self>) -> std::sync::Arc<dyn AnonymousColumnsUdf> {
        Arc::new(Arc::unwrap_or_clone(self))
    }

    fn get_field(
        &self,
        _: &polars::prelude::Schema,
        _: &[polars::prelude::Field],
    ) -> PolarsResult<polars::prelude::Field> {
        Ok(Field::new("len".into(), DataType::UInt32))
    }
}
impl OpenDPPlugin for DPFrameLenShim {
    const NAME: &'static str = "dp_frame_len";
    #[cfg(feature = "ffi")]
    const SHIM: bool = true;
    fn function_options() -> FunctionOptions {
        FunctionOptions::aggregation()
    }
}

pub fn make_expr_dp_frame_len<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([scale]) = match_shim::<DPFrameLenShim, _>(&expr)? else {
        return fallible!(
            MakeMeasurement,
            "Expected {} function",
            DPFrameLenShim::NAME
        );
    };

    apply_plugin(vec![len(), scale], expr, NoiseShim).make_private(
        input_domain.clone(),
        input_metric,
        output_measure,
        global_scale,
    )
}

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