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

# type: ignore
def make_randomized_response(categories: set[T], prob: f64):
    input_domain = AtomDomain(bool)
    input_metric = DiscreteMetric()
    output_measure = MaxDivergence()

    categories = list(categories)

    if len(categories) < 2:  # |\label{line:num_cats}|
        raise ValueError("expected at least two categories")
    
    num_categories = len(categories)

    if not (1 / num_categories <= prob <= 1):  # |\label{line:range}|
        raise ValueError("probability must be within [1/num_categories, 1]")
    
    # prepare constant: |\label{line:map}|
    if prob == 1.0:
        c = float("inf")
    else:
        c = p.inf_div((1).neg_inf_sub(prob)) \
            .inf_mul(num_categories.inf_sub(1)) \
            .inf_ln()
    
    def privacy_map(d_in: u32) -> QO:
        if d_in == 0:
            return 0
        else: 
            return c

    def function(truth: bool) -> bool:  # |\label{line:fn}|
        index = categories.index(truth)
        sample = usize.sample_uniform_int_below(
            len(num_categories) - (0 if index == -1 else 1))
        
        if index != -1 and sample >= index:
            sample += 1
        
        lie = categories[sample]

        be_honest = sample_bernoulli_float(prob, false)
        is_member = index != -1
        return truth if be_honest and is_member else lie
    
    return Measurement(input_domain, function, input_metric, output_measure, privacy_map)